xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs_vnops.c (revision ddb365bfc9e868ad24ccdcb0dc91af18b10df082)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
23  *
24  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
25  *	All rights reserved.
26  */
27 
28 /*
29  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
30  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
31  */
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/time.h>
38 #include <sys/vnode.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/file.h>
42 #include <sys/filio.h>
43 #include <sys/uio.h>
44 #include <sys/buf.h>
45 #include <sys/mman.h>
46 #include <sys/pathname.h>
47 #include <sys/dirent.h>
48 #include <sys/debug.h>
49 #include <sys/vmsystm.h>
50 #include <sys/fcntl.h>
51 #include <sys/flock.h>
52 #include <sys/swap.h>
53 #include <sys/errno.h>
54 #include <sys/strsubr.h>
55 #include <sys/sysmacros.h>
56 #include <sys/kmem.h>
57 #include <sys/cmn_err.h>
58 #include <sys/pathconf.h>
59 #include <sys/utsname.h>
60 #include <sys/dnlc.h>
61 #include <sys/acl.h>
62 #include <sys/atomic.h>
63 #include <sys/policy.h>
64 #include <sys/sdt.h>
65 
66 #include <rpc/types.h>
67 #include <rpc/auth.h>
68 #include <rpc/clnt.h>
69 
70 #include <nfs/nfs.h>
71 #include <nfs/nfs_clnt.h>
72 #include <nfs/rnode.h>
73 #include <nfs/nfs_acl.h>
74 #include <nfs/lm.h>
75 
76 #include <vm/hat.h>
77 #include <vm/as.h>
78 #include <vm/page.h>
79 #include <vm/pvn.h>
80 #include <vm/seg.h>
81 #include <vm/seg_map.h>
82 #include <vm/seg_kpm.h>
83 #include <vm/seg_vn.h>
84 
85 #include <fs/fs_subr.h>
86 
87 #include <sys/ddi.h>
88 
89 static int	nfs_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int,
90 			cred_t *);
91 static int	nfswrite(vnode_t *, caddr_t, uint_t, int, cred_t *);
92 static int	nfsread(vnode_t *, caddr_t, uint_t, int, size_t *, cred_t *);
93 static int	nfssetattr(vnode_t *, struct vattr *, int, cred_t *);
94 static int	nfslookup_dnlc(vnode_t *, char *, vnode_t **, cred_t *);
95 static int	nfslookup_otw(vnode_t *, char *, vnode_t **, cred_t *, int);
96 static int	nfsrename(vnode_t *, char *, vnode_t *, char *, cred_t *,
97 			caller_context_t *);
98 static int	nfsreaddir(vnode_t *, rddir_cache *, cred_t *);
99 static int	nfs_bio(struct buf *, cred_t *);
100 static int	nfs_getapage(vnode_t *, u_offset_t, size_t, uint_t *,
101 			page_t *[], size_t, struct seg *, caddr_t,
102 			enum seg_rw, cred_t *);
103 static void	nfs_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *,
104 			cred_t *);
105 static int	nfs_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t,
106 			int, cred_t *);
107 static int	nfs_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t,
108 			int, cred_t *);
109 static void	nfs_delmap_callback(struct as *, void *, uint_t);
110 
111 /*
112  * Error flags used to pass information about certain special errors
113  * which need to be handled specially.
114  */
115 #define	NFS_EOF			-98
116 
117 /*
118  * These are the vnode ops routines which implement the vnode interface to
119  * the networked file system.  These routines just take their parameters,
120  * make them look networkish by putting the right info into interface structs,
121  * and then calling the appropriate remote routine(s) to do the work.
122  *
123  * Note on directory name lookup cacheing:  If we detect a stale fhandle,
124  * we purge the directory cache relative to that vnode.  This way, the
125  * user won't get burned by the cache repeatedly.  See <nfs/rnode.h> for
126  * more details on rnode locking.
127  */
128 
129 static int	nfs_open(vnode_t **, int, cred_t *, caller_context_t *);
130 static int	nfs_close(vnode_t *, int, int, offset_t, cred_t *,
131 			caller_context_t *);
132 static int	nfs_read(vnode_t *, struct uio *, int, cred_t *,
133 			caller_context_t *);
134 static int	nfs_write(vnode_t *, struct uio *, int, cred_t *,
135 			caller_context_t *);
136 static int	nfs_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
137 			caller_context_t *);
138 static int	nfs_getattr(vnode_t *, struct vattr *, int, cred_t *,
139 			caller_context_t *);
140 static int	nfs_setattr(vnode_t *, struct vattr *, int, cred_t *,
141 			caller_context_t *);
142 static int	nfs_access(vnode_t *, int, int, cred_t *, caller_context_t *);
143 static int	nfs_accessx(void *, int, cred_t *);
144 static int	nfs_readlink(vnode_t *, struct uio *, cred_t *,
145 			caller_context_t *);
146 static int	nfs_fsync(vnode_t *, int, cred_t *, caller_context_t *);
147 static void	nfs_inactive(vnode_t *, cred_t *, caller_context_t *);
148 static int	nfs_lookup(vnode_t *, char *, vnode_t **, struct pathname *,
149 			int, vnode_t *, cred_t *, caller_context_t *,
150 			int *, pathname_t *);
151 static int	nfs_create(vnode_t *, char *, struct vattr *, enum vcexcl,
152 			int, vnode_t **, cred_t *, int, caller_context_t *,
153 			vsecattr_t *);
154 static int	nfs_remove(vnode_t *, char *, cred_t *, caller_context_t *,
155 			int);
156 static int	nfs_link(vnode_t *, vnode_t *, char *, cred_t *,
157 			caller_context_t *, int);
158 static int	nfs_rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
159 			caller_context_t *, int);
160 static int	nfs_mkdir(vnode_t *, char *, struct vattr *, vnode_t **,
161 			cred_t *, caller_context_t *, int, vsecattr_t *);
162 static int	nfs_rmdir(vnode_t *, char *, vnode_t *, cred_t *,
163 			caller_context_t *, int);
164 static int	nfs_symlink(vnode_t *, char *, struct vattr *, char *,
165 			cred_t *, caller_context_t *, int);
166 static int	nfs_readdir(vnode_t *, struct uio *, cred_t *, int *,
167 			caller_context_t *, int);
168 static int	nfs_fid(vnode_t *, fid_t *, caller_context_t *);
169 static int	nfs_rwlock(vnode_t *, int, caller_context_t *);
170 static void	nfs_rwunlock(vnode_t *, int, caller_context_t *);
171 static int	nfs_seek(vnode_t *, offset_t, offset_t *, caller_context_t *);
172 static int	nfs_getpage(vnode_t *, offset_t, size_t, uint_t *,
173 			page_t *[], size_t, struct seg *, caddr_t,
174 			enum seg_rw, cred_t *, caller_context_t *);
175 static int	nfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
176 			caller_context_t *);
177 static int	nfs_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
178 			uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
179 static int	nfs_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
180 			uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
181 static int	nfs_frlock(vnode_t *, int, struct flock64 *, int, offset_t,
182 			struct flk_callback *, cred_t *, caller_context_t *);
183 static int	nfs_space(vnode_t *, int, struct flock64 *, int, offset_t,
184 			cred_t *, caller_context_t *);
185 static int	nfs_realvp(vnode_t *, vnode_t **, caller_context_t *);
186 static int	nfs_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
187 			uint_t, uint_t, uint_t, cred_t *, caller_context_t *);
188 static int	nfs_pathconf(vnode_t *, int, ulong_t *, cred_t *,
189 			caller_context_t *);
190 static int	nfs_pageio(vnode_t *, page_t *, u_offset_t, size_t, int,
191 			cred_t *, caller_context_t *);
192 static int	nfs_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
193 			caller_context_t *);
194 static int	nfs_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
195 			caller_context_t *);
196 static int	nfs_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *,
197 			caller_context_t *);
198 
199 struct vnodeops *nfs_vnodeops;
200 
201 const fs_operation_def_t nfs_vnodeops_template[] = {
202 	VOPNAME_OPEN,		{ .vop_open = nfs_open },
203 	VOPNAME_CLOSE,		{ .vop_close = nfs_close },
204 	VOPNAME_READ,		{ .vop_read = nfs_read },
205 	VOPNAME_WRITE,		{ .vop_write = nfs_write },
206 	VOPNAME_IOCTL,		{ .vop_ioctl = nfs_ioctl },
207 	VOPNAME_GETATTR,	{ .vop_getattr = nfs_getattr },
208 	VOPNAME_SETATTR,	{ .vop_setattr = nfs_setattr },
209 	VOPNAME_ACCESS,		{ .vop_access = nfs_access },
210 	VOPNAME_LOOKUP,		{ .vop_lookup = nfs_lookup },
211 	VOPNAME_CREATE,		{ .vop_create = nfs_create },
212 	VOPNAME_REMOVE,		{ .vop_remove = nfs_remove },
213 	VOPNAME_LINK,		{ .vop_link = nfs_link },
214 	VOPNAME_RENAME,		{ .vop_rename = nfs_rename },
215 	VOPNAME_MKDIR,		{ .vop_mkdir = nfs_mkdir },
216 	VOPNAME_RMDIR,		{ .vop_rmdir = nfs_rmdir },
217 	VOPNAME_READDIR,	{ .vop_readdir = nfs_readdir },
218 	VOPNAME_SYMLINK,	{ .vop_symlink = nfs_symlink },
219 	VOPNAME_READLINK,	{ .vop_readlink = nfs_readlink },
220 	VOPNAME_FSYNC,		{ .vop_fsync = nfs_fsync },
221 	VOPNAME_INACTIVE,	{ .vop_inactive = nfs_inactive },
222 	VOPNAME_FID,		{ .vop_fid = nfs_fid },
223 	VOPNAME_RWLOCK,		{ .vop_rwlock = nfs_rwlock },
224 	VOPNAME_RWUNLOCK,	{ .vop_rwunlock = nfs_rwunlock },
225 	VOPNAME_SEEK,		{ .vop_seek = nfs_seek },
226 	VOPNAME_FRLOCK,		{ .vop_frlock = nfs_frlock },
227 	VOPNAME_SPACE,		{ .vop_space = nfs_space },
228 	VOPNAME_REALVP,		{ .vop_realvp = nfs_realvp },
229 	VOPNAME_GETPAGE,	{ .vop_getpage = nfs_getpage },
230 	VOPNAME_PUTPAGE,	{ .vop_putpage = nfs_putpage },
231 	VOPNAME_MAP,		{ .vop_map = nfs_map },
232 	VOPNAME_ADDMAP,		{ .vop_addmap = nfs_addmap },
233 	VOPNAME_DELMAP,		{ .vop_delmap = nfs_delmap },
234 	VOPNAME_DUMP,		{ .vop_dump = nfs_dump },
235 	VOPNAME_PATHCONF,	{ .vop_pathconf = nfs_pathconf },
236 	VOPNAME_PAGEIO,		{ .vop_pageio = nfs_pageio },
237 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = nfs_setsecattr },
238 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = nfs_getsecattr },
239 	VOPNAME_SHRLOCK,	{ .vop_shrlock = nfs_shrlock },
240 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
241 	NULL,			NULL
242 };
243 
244 /*
245  * XXX:  This is referenced in modstubs.S
246  */
247 struct vnodeops *
248 nfs_getvnodeops(void)
249 {
250 	return (nfs_vnodeops);
251 }
252 
253 /* ARGSUSED */
254 static int
255 nfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
256 {
257 	int error;
258 	struct vattr va;
259 	rnode_t *rp;
260 	vnode_t *vp;
261 
262 	vp = *vpp;
263 	rp = VTOR(vp);
264 	if (nfs_zone() != VTOMI(vp)->mi_zone)
265 		return (EIO);
266 	mutex_enter(&rp->r_statelock);
267 	if (rp->r_cred == NULL) {
268 		crhold(cr);
269 		rp->r_cred = cr;
270 	}
271 	mutex_exit(&rp->r_statelock);
272 
273 	/*
274 	 * If there is no cached data or if close-to-open
275 	 * consistency checking is turned off, we can avoid
276 	 * the over the wire getattr.  Otherwise, if the
277 	 * file system is mounted readonly, then just verify
278 	 * the caches are up to date using the normal mechanism.
279 	 * Else, if the file is not mmap'd, then just mark
280 	 * the attributes as timed out.  They will be refreshed
281 	 * and the caches validated prior to being used.
282 	 * Else, the file system is mounted writeable so
283 	 * force an over the wire GETATTR in order to ensure
284 	 * that all cached data is valid.
285 	 */
286 	if (vp->v_count > 1 ||
287 	    ((vn_has_cached_data(vp) || HAVE_RDDIR_CACHE(rp)) &&
288 	    !(VTOMI(vp)->mi_flags & MI_NOCTO))) {
289 		if (vn_is_readonly(vp))
290 			error = nfs_validate_caches(vp, cr);
291 		else if (rp->r_mapcnt == 0 && vp->v_count == 1) {
292 			PURGE_ATTRCACHE(vp);
293 			error = 0;
294 		} else {
295 			va.va_mask = AT_ALL;
296 			error = nfs_getattr_otw(vp, &va, cr);
297 		}
298 	} else
299 		error = 0;
300 
301 	return (error);
302 }
303 
304 /* ARGSUSED */
305 static int
306 nfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
307 	caller_context_t *ct)
308 {
309 	rnode_t *rp;
310 	int error;
311 	struct vattr va;
312 
313 	/*
314 	 * zone_enter(2) prevents processes from changing zones with NFS files
315 	 * open; if we happen to get here from the wrong zone we can't do
316 	 * anything over the wire.
317 	 */
318 	if (VTOMI(vp)->mi_zone != nfs_zone()) {
319 		/*
320 		 * We could attempt to clean up locks, except we're sure
321 		 * that the current process didn't acquire any locks on
322 		 * the file: any attempt to lock a file belong to another zone
323 		 * will fail, and one can't lock an NFS file and then change
324 		 * zones, as that fails too.
325 		 *
326 		 * Returning an error here is the sane thing to do.  A
327 		 * subsequent call to VN_RELE() which translates to a
328 		 * nfs_inactive() will clean up state: if the zone of the
329 		 * vnode's origin is still alive and kicking, an async worker
330 		 * thread will handle the request (from the correct zone), and
331 		 * everything (minus the final nfs_getattr_otw() call) should
332 		 * be OK. If the zone is going away nfs_async_inactive() will
333 		 * throw away cached pages inline.
334 		 */
335 		return (EIO);
336 	}
337 
338 	/*
339 	 * If we are using local locking for this filesystem, then
340 	 * release all of the SYSV style record locks.  Otherwise,
341 	 * we are doing network locking and we need to release all
342 	 * of the network locks.  All of the locks held by this
343 	 * process on this file are released no matter what the
344 	 * incoming reference count is.
345 	 */
346 	if (VTOMI(vp)->mi_flags & MI_LLOCK) {
347 		cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
348 		cleanshares(vp, ttoproc(curthread)->p_pid);
349 	} else
350 		nfs_lockrelease(vp, flag, offset, cr);
351 
352 	if (count > 1)
353 		return (0);
354 
355 	/*
356 	 * If the file has been `unlinked', then purge the
357 	 * DNLC so that this vnode will get reycled quicker
358 	 * and the .nfs* file on the server will get removed.
359 	 */
360 	rp = VTOR(vp);
361 	if (rp->r_unldvp != NULL)
362 		dnlc_purge_vp(vp);
363 
364 	/*
365 	 * If the file was open for write and there are pages,
366 	 * then if the file system was mounted using the "no-close-
367 	 *	to-open" semantics, then start an asynchronous flush
368 	 *	of the all of the pages in the file.
369 	 * else the file system was not mounted using the "no-close-
370 	 *	to-open" semantics, then do a synchronous flush and
371 	 *	commit of all of the dirty and uncommitted pages.
372 	 *
373 	 * The asynchronous flush of the pages in the "nocto" path
374 	 * mostly just associates a cred pointer with the rnode so
375 	 * writes which happen later will have a better chance of
376 	 * working.  It also starts the data being written to the
377 	 * server, but without unnecessarily delaying the application.
378 	 */
379 	if ((flag & FWRITE) && vn_has_cached_data(vp)) {
380 		if ((VTOMI(vp)->mi_flags & MI_NOCTO)) {
381 			error = nfs_putpage(vp, (offset_t)0, 0, B_ASYNC,
382 			    cr, ct);
383 			if (error == EAGAIN)
384 				error = 0;
385 		} else
386 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
387 		if (!error) {
388 			mutex_enter(&rp->r_statelock);
389 			error = rp->r_error;
390 			rp->r_error = 0;
391 			mutex_exit(&rp->r_statelock);
392 		}
393 	} else {
394 		mutex_enter(&rp->r_statelock);
395 		error = rp->r_error;
396 		rp->r_error = 0;
397 		mutex_exit(&rp->r_statelock);
398 	}
399 
400 	/*
401 	 * If RWRITEATTR is set, then issue an over the wire GETATTR to
402 	 * refresh the attribute cache with a set of attributes which
403 	 * weren't returned from a WRITE.  This will enable the close-
404 	 * to-open processing to work.
405 	 */
406 	if (rp->r_flags & RWRITEATTR)
407 		(void) nfs_getattr_otw(vp, &va, cr);
408 
409 	return (error);
410 }
411 
412 /* ARGSUSED */
413 static int
414 nfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
415 	caller_context_t *ct)
416 {
417 	rnode_t *rp;
418 	u_offset_t off;
419 	offset_t diff;
420 	int on;
421 	size_t n;
422 	caddr_t base;
423 	uint_t flags;
424 	int error;
425 	mntinfo_t *mi;
426 
427 	rp = VTOR(vp);
428 	mi = VTOMI(vp);
429 
430 	if (nfs_zone() != mi->mi_zone)
431 		return (EIO);
432 
433 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
434 
435 	if (vp->v_type != VREG)
436 		return (EISDIR);
437 
438 	if (uiop->uio_resid == 0)
439 		return (0);
440 
441 	if (uiop->uio_loffset > MAXOFF32_T)
442 		return (EFBIG);
443 
444 	if (uiop->uio_loffset < 0 ||
445 	    uiop->uio_loffset + uiop->uio_resid > MAXOFF32_T)
446 		return (EINVAL);
447 
448 	/*
449 	 * Bypass VM if caching has been disabled (e.g., locking) or if
450 	 * using client-side direct I/O and the file is not mmap'd and
451 	 * there are no cached pages.
452 	 */
453 	if ((vp->v_flag & VNOCACHE) ||
454 	    (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
455 	    rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
456 	    !vn_has_cached_data(vp))) {
457 		size_t bufsize;
458 		size_t resid = 0;
459 
460 		/*
461 		 * Let's try to do read in as large a chunk as we can
462 		 * (Filesystem (NFS client) bsize if possible/needed).
463 		 * For V3, this is 32K and for V2, this is 8K.
464 		 */
465 		bufsize = MIN(uiop->uio_resid, VTOMI(vp)->mi_curread);
466 		base = kmem_alloc(bufsize, KM_SLEEP);
467 		do {
468 			n = MIN(uiop->uio_resid, bufsize);
469 			error = nfsread(vp, base, uiop->uio_offset, n,
470 			    &resid, cr);
471 			if (!error) {
472 				n -= resid;
473 				error = uiomove(base, n, UIO_READ, uiop);
474 			}
475 		} while (!error && uiop->uio_resid > 0 && n > 0);
476 		kmem_free(base, bufsize);
477 		return (error);
478 	}
479 
480 	error = 0;
481 
482 	do {
483 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
484 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
485 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
486 
487 		error = nfs_validate_caches(vp, cr);
488 		if (error)
489 			break;
490 
491 		mutex_enter(&rp->r_statelock);
492 		while (rp->r_flags & RINCACHEPURGE) {
493 			if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
494 				mutex_exit(&rp->r_statelock);
495 				return (EINTR);
496 			}
497 		}
498 		diff = rp->r_size - uiop->uio_loffset;
499 		mutex_exit(&rp->r_statelock);
500 		if (diff <= 0)
501 			break;
502 		if (diff < n)
503 			n = (size_t)diff;
504 
505 		if (vpm_enable) {
506 			/*
507 			 * Copy data.
508 			 */
509 			error = vpm_data_copy(vp, off + on, n, uiop,
510 			    1, NULL, 0, S_READ);
511 		} else {
512 			base = segmap_getmapflt(segkmap, vp, off + on, n,
513 			    1, S_READ);
514 			error = uiomove(base + on, n, UIO_READ, uiop);
515 		}
516 
517 		if (!error) {
518 			/*
519 			 * If read a whole block or read to eof,
520 			 * won't need this buffer again soon.
521 			 */
522 			mutex_enter(&rp->r_statelock);
523 			if (n + on == MAXBSIZE ||
524 			    uiop->uio_loffset == rp->r_size)
525 				flags = SM_DONTNEED;
526 			else
527 				flags = 0;
528 			mutex_exit(&rp->r_statelock);
529 			if (vpm_enable) {
530 				error = vpm_sync_pages(vp, off, n, flags);
531 			} else {
532 				error = segmap_release(segkmap, base, flags);
533 			}
534 		} else {
535 			if (vpm_enable) {
536 				(void) vpm_sync_pages(vp, off, n, 0);
537 			} else {
538 				(void) segmap_release(segkmap, base, 0);
539 			}
540 		}
541 	} while (!error && uiop->uio_resid > 0);
542 
543 	return (error);
544 }
545 
546 /* ARGSUSED */
547 static int
548 nfs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
549 	caller_context_t *ct)
550 {
551 	rnode_t *rp;
552 	u_offset_t off;
553 	caddr_t base;
554 	uint_t flags;
555 	int remainder;
556 	size_t n;
557 	int on;
558 	int error;
559 	int resid;
560 	offset_t offset;
561 	rlim_t limit;
562 	mntinfo_t *mi;
563 
564 	rp = VTOR(vp);
565 
566 	mi = VTOMI(vp);
567 	if (nfs_zone() != mi->mi_zone)
568 		return (EIO);
569 	if (vp->v_type != VREG)
570 		return (EISDIR);
571 
572 	if (uiop->uio_resid == 0)
573 		return (0);
574 
575 	if (ioflag & FAPPEND) {
576 		struct vattr va;
577 
578 		/*
579 		 * Must serialize if appending.
580 		 */
581 		if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) {
582 			nfs_rw_exit(&rp->r_rwlock);
583 			if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER,
584 			    INTR(vp)))
585 				return (EINTR);
586 		}
587 
588 		va.va_mask = AT_SIZE;
589 		error = nfsgetattr(vp, &va, cr);
590 		if (error)
591 			return (error);
592 		uiop->uio_loffset = va.va_size;
593 	}
594 
595 	if (uiop->uio_loffset > MAXOFF32_T)
596 		return (EFBIG);
597 
598 	offset = uiop->uio_loffset + uiop->uio_resid;
599 
600 	if (uiop->uio_loffset < 0 || offset > MAXOFF32_T)
601 		return (EINVAL);
602 
603 	if (uiop->uio_llimit > (rlim64_t)MAXOFF32_T) {
604 		limit = MAXOFF32_T;
605 	} else {
606 		limit = (rlim_t)uiop->uio_llimit;
607 	}
608 
609 	/*
610 	 * Check to make sure that the process will not exceed
611 	 * its limit on file size.  It is okay to write up to
612 	 * the limit, but not beyond.  Thus, the write which
613 	 * reaches the limit will be short and the next write
614 	 * will return an error.
615 	 */
616 	remainder = 0;
617 	if (offset > limit) {
618 		remainder = offset - limit;
619 		uiop->uio_resid = limit - uiop->uio_offset;
620 		if (uiop->uio_resid <= 0) {
621 			proc_t *p = ttoproc(curthread);
622 
623 			uiop->uio_resid += remainder;
624 			mutex_enter(&p->p_lock);
625 			(void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE],
626 			    p->p_rctls, p, RCA_UNSAFE_SIGINFO);
627 			mutex_exit(&p->p_lock);
628 			return (EFBIG);
629 		}
630 	}
631 
632 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp)))
633 		return (EINTR);
634 
635 	/*
636 	 * Bypass VM if caching has been disabled (e.g., locking) or if
637 	 * using client-side direct I/O and the file is not mmap'd and
638 	 * there are no cached pages.
639 	 */
640 	if ((vp->v_flag & VNOCACHE) ||
641 	    (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
642 	    rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
643 	    !vn_has_cached_data(vp))) {
644 		size_t bufsize;
645 		int count;
646 		uint_t org_offset;
647 
648 nfs_fwrite:
649 		if (rp->r_flags & RSTALE) {
650 			resid = uiop->uio_resid;
651 			offset = uiop->uio_loffset;
652 			error = rp->r_error;
653 			/*
654 			 * A close may have cleared r_error, if so,
655 			 * propagate ESTALE error return properly
656 			 */
657 			if (error == 0)
658 				error = ESTALE;
659 			goto bottom;
660 		}
661 		bufsize = MIN(uiop->uio_resid, mi->mi_curwrite);
662 		base = kmem_alloc(bufsize, KM_SLEEP);
663 		do {
664 			resid = uiop->uio_resid;
665 			offset = uiop->uio_loffset;
666 			count = MIN(uiop->uio_resid, bufsize);
667 			org_offset = uiop->uio_offset;
668 			error = uiomove(base, count, UIO_WRITE, uiop);
669 			if (!error) {
670 				error = nfswrite(vp, base, org_offset,
671 				    count, cr);
672 			}
673 		} while (!error && uiop->uio_resid > 0);
674 		kmem_free(base, bufsize);
675 		goto bottom;
676 	}
677 
678 	do {
679 		off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
680 		on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
681 		n = MIN(MAXBSIZE - on, uiop->uio_resid);
682 
683 		resid = uiop->uio_resid;
684 		offset = uiop->uio_loffset;
685 
686 		if (rp->r_flags & RSTALE) {
687 			error = rp->r_error;
688 			/*
689 			 * A close may have cleared r_error, if so,
690 			 * propagate ESTALE error return properly
691 			 */
692 			if (error == 0)
693 				error = ESTALE;
694 			break;
695 		}
696 
697 		/*
698 		 * Don't create dirty pages faster than they
699 		 * can be cleaned so that the system doesn't
700 		 * get imbalanced.  If the async queue is
701 		 * maxed out, then wait for it to drain before
702 		 * creating more dirty pages.  Also, wait for
703 		 * any threads doing pagewalks in the vop_getattr
704 		 * entry points so that they don't block for
705 		 * long periods.
706 		 */
707 		mutex_enter(&rp->r_statelock);
708 		while ((mi->mi_max_threads != 0 &&
709 		    rp->r_awcount > 2 * mi->mi_max_threads) ||
710 		    rp->r_gcount > 0) {
711 			if (INTR(vp)) {
712 				klwp_t *lwp = ttolwp(curthread);
713 
714 				if (lwp != NULL)
715 					lwp->lwp_nostop++;
716 				if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
717 					mutex_exit(&rp->r_statelock);
718 					if (lwp != NULL)
719 						lwp->lwp_nostop--;
720 					error = EINTR;
721 					goto bottom;
722 				}
723 				if (lwp != NULL)
724 					lwp->lwp_nostop--;
725 			} else
726 				cv_wait(&rp->r_cv, &rp->r_statelock);
727 		}
728 		mutex_exit(&rp->r_statelock);
729 
730 		/*
731 		 * Touch the page and fault it in if it is not in core
732 		 * before segmap_getmapflt or vpm_data_copy can lock it.
733 		 * This is to avoid the deadlock if the buffer is mapped
734 		 * to the same file through mmap which we want to write.
735 		 */
736 		uio_prefaultpages((long)n, uiop);
737 
738 		if (vpm_enable) {
739 			/*
740 			 * It will use kpm mappings, so no need to
741 			 * pass an address.
742 			 */
743 			error = writerp(rp, NULL, n, uiop, 0);
744 		} else  {
745 			if (segmap_kpm) {
746 				int pon = uiop->uio_loffset & PAGEOFFSET;
747 				size_t pn = MIN(PAGESIZE - pon,
748 				    uiop->uio_resid);
749 				int pagecreate;
750 
751 				mutex_enter(&rp->r_statelock);
752 				pagecreate = (pon == 0) && (pn == PAGESIZE ||
753 				    uiop->uio_loffset + pn >= rp->r_size);
754 				mutex_exit(&rp->r_statelock);
755 
756 				base = segmap_getmapflt(segkmap, vp, off + on,
757 				    pn, !pagecreate, S_WRITE);
758 
759 				error = writerp(rp, base + pon, n, uiop,
760 				    pagecreate);
761 
762 			} else {
763 				base = segmap_getmapflt(segkmap, vp, off + on,
764 				    n, 0, S_READ);
765 				error = writerp(rp, base + on, n, uiop, 0);
766 			}
767 		}
768 
769 		if (!error) {
770 			if (mi->mi_flags & MI_NOAC)
771 				flags = SM_WRITE;
772 			else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
773 				/*
774 				 * Have written a whole block.
775 				 * Start an asynchronous write
776 				 * and mark the buffer to
777 				 * indicate that it won't be
778 				 * needed again soon.
779 				 */
780 				flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
781 			} else
782 				flags = 0;
783 			if ((ioflag & (FSYNC|FDSYNC)) ||
784 			    (rp->r_flags & ROUTOFSPACE)) {
785 				flags &= ~SM_ASYNC;
786 				flags |= SM_WRITE;
787 			}
788 			if (vpm_enable) {
789 				error = vpm_sync_pages(vp, off, n, flags);
790 			} else {
791 				error = segmap_release(segkmap, base, flags);
792 			}
793 		} else {
794 			if (vpm_enable) {
795 				(void) vpm_sync_pages(vp, off, n, 0);
796 			} else {
797 				(void) segmap_release(segkmap, base, 0);
798 			}
799 			/*
800 			 * In the event that we got an access error while
801 			 * faulting in a page for a write-only file just
802 			 * force a write.
803 			 */
804 			if (error == EACCES)
805 				goto nfs_fwrite;
806 		}
807 	} while (!error && uiop->uio_resid > 0);
808 
809 bottom:
810 	if (error) {
811 		uiop->uio_resid = resid + remainder;
812 		uiop->uio_loffset = offset;
813 	} else
814 		uiop->uio_resid += remainder;
815 
816 	nfs_rw_exit(&rp->r_lkserlock);
817 
818 	return (error);
819 }
820 
821 /*
822  * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
823  */
824 static int
825 nfs_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
826 	int flags, cred_t *cr)
827 {
828 	struct buf *bp;
829 	int error;
830 
831 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
832 	bp = pageio_setup(pp, len, vp, flags);
833 	ASSERT(bp != NULL);
834 
835 	/*
836 	 * pageio_setup should have set b_addr to 0.  This
837 	 * is correct since we want to do I/O on a page
838 	 * boundary.  bp_mapin will use this addr to calculate
839 	 * an offset, and then set b_addr to the kernel virtual
840 	 * address it allocated for us.
841 	 */
842 	ASSERT(bp->b_un.b_addr == 0);
843 
844 	bp->b_edev = 0;
845 	bp->b_dev = 0;
846 	bp->b_lblkno = lbtodb(off);
847 	bp->b_file = vp;
848 	bp->b_offset = (offset_t)off;
849 	bp_mapin(bp);
850 
851 	error = nfs_bio(bp, cr);
852 
853 	bp_mapout(bp);
854 	pageio_done(bp);
855 
856 	return (error);
857 }
858 
859 /*
860  * Write to file.  Writes to remote server in largest size
861  * chunks that the server can handle.  Write is synchronous.
862  */
863 static int
864 nfswrite(vnode_t *vp, caddr_t base, uint_t offset, int count, cred_t *cr)
865 {
866 	rnode_t *rp;
867 	mntinfo_t *mi;
868 	struct nfswriteargs wa;
869 	struct nfsattrstat ns;
870 	int error;
871 	int tsize;
872 	int douprintf;
873 
874 	douprintf = 1;
875 
876 	rp = VTOR(vp);
877 	mi = VTOMI(vp);
878 
879 	ASSERT(nfs_zone() == mi->mi_zone);
880 
881 	wa.wa_args = &wa.wa_args_buf;
882 	wa.wa_fhandle = *VTOFH(vp);
883 
884 	do {
885 		tsize = MIN(mi->mi_curwrite, count);
886 		wa.wa_data = base;
887 		wa.wa_begoff = offset;
888 		wa.wa_totcount = tsize;
889 		wa.wa_count = tsize;
890 		wa.wa_offset = offset;
891 
892 		if (mi->mi_io_kstats) {
893 			mutex_enter(&mi->mi_lock);
894 			kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
895 			mutex_exit(&mi->mi_lock);
896 		}
897 		wa.wa_mblk = NULL;
898 		do {
899 			error = rfs2call(mi, RFS_WRITE,
900 			    xdr_writeargs, (caddr_t)&wa,
901 			    xdr_attrstat, (caddr_t)&ns, cr,
902 			    &douprintf, &ns.ns_status, 0, NULL);
903 		} while (error == ENFS_TRYAGAIN);
904 		if (mi->mi_io_kstats) {
905 			mutex_enter(&mi->mi_lock);
906 			kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
907 			mutex_exit(&mi->mi_lock);
908 		}
909 
910 		if (!error) {
911 			error = geterrno(ns.ns_status);
912 			/*
913 			 * Can't check for stale fhandle and purge caches
914 			 * here because pages are held by nfs_getpage.
915 			 * Just mark the attribute cache as timed out
916 			 * and set RWRITEATTR to indicate that the file
917 			 * was modified with a WRITE operation.
918 			 */
919 			if (!error) {
920 				count -= tsize;
921 				base += tsize;
922 				offset += tsize;
923 				if (mi->mi_io_kstats) {
924 					mutex_enter(&mi->mi_lock);
925 					KSTAT_IO_PTR(mi->mi_io_kstats)->
926 					    writes++;
927 					KSTAT_IO_PTR(mi->mi_io_kstats)->
928 					    nwritten += tsize;
929 					mutex_exit(&mi->mi_lock);
930 				}
931 				lwp_stat_update(LWP_STAT_OUBLK, 1);
932 				mutex_enter(&rp->r_statelock);
933 				PURGE_ATTRCACHE_LOCKED(rp);
934 				rp->r_flags |= RWRITEATTR;
935 				mutex_exit(&rp->r_statelock);
936 			}
937 		}
938 	} while (!error && count);
939 
940 	return (error);
941 }
942 
943 /*
944  * Read from a file.  Reads data in largest chunks our interface can handle.
945  */
946 static int
947 nfsread(vnode_t *vp, caddr_t base, uint_t offset,
948     int count, size_t *residp, cred_t *cr)
949 {
950 	mntinfo_t *mi;
951 	struct nfsreadargs ra;
952 	struct nfsrdresult rr;
953 	int tsize;
954 	int error;
955 	int douprintf;
956 	failinfo_t fi;
957 	rnode_t *rp;
958 	struct vattr va;
959 	hrtime_t t;
960 
961 	rp = VTOR(vp);
962 	mi = VTOMI(vp);
963 
964 	ASSERT(nfs_zone() == mi->mi_zone);
965 
966 	douprintf = 1;
967 
968 	ra.ra_fhandle = *VTOFH(vp);
969 
970 	fi.vp = vp;
971 	fi.fhp = (caddr_t)&ra.ra_fhandle;
972 	fi.copyproc = nfscopyfh;
973 	fi.lookupproc = nfslookup;
974 	fi.xattrdirproc = acl_getxattrdir2;
975 
976 	do {
977 		if (mi->mi_io_kstats) {
978 			mutex_enter(&mi->mi_lock);
979 			kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
980 			mutex_exit(&mi->mi_lock);
981 		}
982 
983 		do {
984 			tsize = MIN(mi->mi_curread, count);
985 			rr.rr_data = base;
986 			ra.ra_offset = offset;
987 			ra.ra_totcount = tsize;
988 			ra.ra_count = tsize;
989 			ra.ra_data = base;
990 			t = gethrtime();
991 			error = rfs2call(mi, RFS_READ,
992 			    xdr_readargs, (caddr_t)&ra,
993 			    xdr_rdresult, (caddr_t)&rr, cr,
994 			    &douprintf, &rr.rr_status, 0, &fi);
995 		} while (error == ENFS_TRYAGAIN);
996 
997 		if (mi->mi_io_kstats) {
998 			mutex_enter(&mi->mi_lock);
999 			kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
1000 			mutex_exit(&mi->mi_lock);
1001 		}
1002 
1003 		if (!error) {
1004 			error = geterrno(rr.rr_status);
1005 			if (!error) {
1006 				count -= rr.rr_count;
1007 				base += rr.rr_count;
1008 				offset += rr.rr_count;
1009 				if (mi->mi_io_kstats) {
1010 					mutex_enter(&mi->mi_lock);
1011 					KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
1012 					KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
1013 					    rr.rr_count;
1014 					mutex_exit(&mi->mi_lock);
1015 				}
1016 				lwp_stat_update(LWP_STAT_INBLK, 1);
1017 			}
1018 		}
1019 	} while (!error && count && rr.rr_count == tsize);
1020 
1021 	*residp = count;
1022 
1023 	if (!error) {
1024 		/*
1025 		 * Since no error occurred, we have the current
1026 		 * attributes and we need to do a cache check and then
1027 		 * potentially update the cached attributes.  We can't
1028 		 * use the normal attribute check and cache mechanisms
1029 		 * because they might cause a cache flush which would
1030 		 * deadlock.  Instead, we just check the cache to see
1031 		 * if the attributes have changed.  If it is, then we
1032 		 * just mark the attributes as out of date.  The next
1033 		 * time that the attributes are checked, they will be
1034 		 * out of date, new attributes will be fetched, and
1035 		 * the page cache will be flushed.  If the attributes
1036 		 * weren't changed, then we just update the cached
1037 		 * attributes with these attributes.
1038 		 */
1039 		/*
1040 		 * If NFS_ACL is supported on the server, then the
1041 		 * attributes returned by server may have minimal
1042 		 * permissions sometimes denying access to users having
1043 		 * proper access.  To get the proper attributes, mark
1044 		 * the attributes as expired so that they will be
1045 		 * regotten via the NFS_ACL GETATTR2 procedure.
1046 		 */
1047 		error = nattr_to_vattr(vp, &rr.rr_attr, &va);
1048 		mutex_enter(&rp->r_statelock);
1049 		if (error || !CACHE_VALID(rp, va.va_mtime, va.va_size) ||
1050 		    (mi->mi_flags & MI_ACL)) {
1051 			mutex_exit(&rp->r_statelock);
1052 			PURGE_ATTRCACHE(vp);
1053 		} else {
1054 			if (rp->r_mtime <= t) {
1055 				nfs_attrcache_va(vp, &va);
1056 			}
1057 			mutex_exit(&rp->r_statelock);
1058 		}
1059 	}
1060 
1061 	return (error);
1062 }
1063 
1064 /* ARGSUSED */
1065 static int
1066 nfs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp,
1067 	caller_context_t *ct)
1068 {
1069 
1070 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1071 		return (EIO);
1072 	switch (cmd) {
1073 		case _FIODIRECTIO:
1074 			return (nfs_directio(vp, (int)arg, cr));
1075 		default:
1076 			return (ENOTTY);
1077 	}
1078 }
1079 
1080 /* ARGSUSED */
1081 static int
1082 nfs_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1083 	caller_context_t *ct)
1084 {
1085 	int error;
1086 	rnode_t *rp;
1087 
1088 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1089 		return (EIO);
1090 	/*
1091 	 * If it has been specified that the return value will
1092 	 * just be used as a hint, and we are only being asked
1093 	 * for size, fsid or rdevid, then return the client's
1094 	 * notion of these values without checking to make sure
1095 	 * that the attribute cache is up to date.
1096 	 * The whole point is to avoid an over the wire GETATTR
1097 	 * call.
1098 	 */
1099 	rp = VTOR(vp);
1100 	if (flags & ATTR_HINT) {
1101 		if (vap->va_mask ==
1102 		    (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) {
1103 			mutex_enter(&rp->r_statelock);
1104 			if (vap->va_mask | AT_SIZE)
1105 				vap->va_size = rp->r_size;
1106 			if (vap->va_mask | AT_FSID)
1107 				vap->va_fsid = rp->r_attr.va_fsid;
1108 			if (vap->va_mask | AT_RDEV)
1109 				vap->va_rdev = rp->r_attr.va_rdev;
1110 			mutex_exit(&rp->r_statelock);
1111 			return (0);
1112 		}
1113 	}
1114 
1115 	/*
1116 	 * Only need to flush pages if asking for the mtime
1117 	 * and if there any dirty pages or any outstanding
1118 	 * asynchronous (write) requests for this file.
1119 	 */
1120 	if (vap->va_mask & AT_MTIME) {
1121 		if (vn_has_cached_data(vp) &&
1122 		    ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) {
1123 			mutex_enter(&rp->r_statelock);
1124 			rp->r_gcount++;
1125 			mutex_exit(&rp->r_statelock);
1126 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1127 			mutex_enter(&rp->r_statelock);
1128 			if (error && (error == ENOSPC || error == EDQUOT)) {
1129 				if (!rp->r_error)
1130 					rp->r_error = error;
1131 			}
1132 			if (--rp->r_gcount == 0)
1133 				cv_broadcast(&rp->r_cv);
1134 			mutex_exit(&rp->r_statelock);
1135 		}
1136 	}
1137 
1138 	return (nfsgetattr(vp, vap, cr));
1139 }
1140 
1141 /*ARGSUSED4*/
1142 static int
1143 nfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1144 		caller_context_t *ct)
1145 {
1146 	int error;
1147 	uint_t mask;
1148 	struct vattr va;
1149 
1150 	mask = vap->va_mask;
1151 
1152 	if (mask & AT_NOSET)
1153 		return (EINVAL);
1154 
1155 	if ((mask & AT_SIZE) &&
1156 	    vap->va_type == VREG &&
1157 	    vap->va_size > MAXOFF32_T)
1158 		return (EFBIG);
1159 
1160 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1161 		return (EIO);
1162 
1163 	va.va_mask = AT_UID | AT_MODE;
1164 
1165 	error = nfsgetattr(vp, &va, cr);
1166 	if (error)
1167 		return (error);
1168 
1169 	error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs_accessx,
1170 	    vp);
1171 
1172 	if (error)
1173 		return (error);
1174 
1175 	error = nfssetattr(vp, vap, flags, cr);
1176 
1177 	if (error == 0 && (mask & AT_SIZE) && vap->va_size == 0)
1178 		vnevent_truncate(vp, ct);
1179 
1180 	return (error);
1181 }
1182 
1183 static int
1184 nfssetattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr)
1185 {
1186 	int error;
1187 	uint_t mask;
1188 	struct nfssaargs args;
1189 	struct nfsattrstat ns;
1190 	int douprintf;
1191 	rnode_t *rp;
1192 	struct vattr va;
1193 	mode_t omode;
1194 	mntinfo_t *mi;
1195 	vsecattr_t *vsp;
1196 	hrtime_t t;
1197 
1198 	mask = vap->va_mask;
1199 
1200 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
1201 
1202 	rp = VTOR(vp);
1203 
1204 	/*
1205 	 * Only need to flush pages if there are any pages and
1206 	 * if the file is marked as dirty in some fashion.  The
1207 	 * file must be flushed so that we can accurately
1208 	 * determine the size of the file and the cached data
1209 	 * after the SETATTR returns.  A file is considered to
1210 	 * be dirty if it is either marked with RDIRTY, has
1211 	 * outstanding i/o's active, or is mmap'd.  In this
1212 	 * last case, we can't tell whether there are dirty
1213 	 * pages, so we flush just to be sure.
1214 	 */
1215 	if (vn_has_cached_data(vp) &&
1216 	    ((rp->r_flags & RDIRTY) ||
1217 	    rp->r_count > 0 ||
1218 	    rp->r_mapcnt > 0)) {
1219 		ASSERT(vp->v_type != VCHR);
1220 		error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, NULL);
1221 		if (error && (error == ENOSPC || error == EDQUOT)) {
1222 			mutex_enter(&rp->r_statelock);
1223 			if (!rp->r_error)
1224 				rp->r_error = error;
1225 			mutex_exit(&rp->r_statelock);
1226 		}
1227 	}
1228 
1229 	/*
1230 	 * If the system call was utime(2) or utimes(2) and the
1231 	 * application did not specify the times, then set the
1232 	 * mtime nanosecond field to 1 billion.  This will get
1233 	 * translated from 1 billion nanoseconds to 1 million
1234 	 * microseconds in the over the wire request.  The
1235 	 * server will use 1 million in the microsecond field
1236 	 * to tell whether both the mtime and atime should be
1237 	 * set to the server's current time.
1238 	 *
1239 	 * This is an overload of the protocol and should be
1240 	 * documented in the NFS Version 2 protocol specification.
1241 	 */
1242 	if ((mask & AT_MTIME) && !(flags & ATTR_UTIME)) {
1243 		vap->va_mtime.tv_nsec = 1000000000;
1244 		if (NFS_TIME_T_OK(vap->va_mtime.tv_sec) &&
1245 		    NFS_TIME_T_OK(vap->va_atime.tv_sec)) {
1246 			error = vattr_to_sattr(vap, &args.saa_sa);
1247 		} else {
1248 			/*
1249 			 * Use server times. vap time values will not be used.
1250 			 * To ensure no time overflow, make sure vap has
1251 			 * valid values, but retain the original values.
1252 			 */
1253 			timestruc_t	mtime = vap->va_mtime;
1254 			timestruc_t	atime = vap->va_atime;
1255 			time_t		now;
1256 
1257 			now = gethrestime_sec();
1258 			if (NFS_TIME_T_OK(now)) {
1259 				/* Just in case server does not know of this */
1260 				vap->va_mtime.tv_sec = now;
1261 				vap->va_atime.tv_sec = now;
1262 			} else {
1263 				vap->va_mtime.tv_sec = 0;
1264 				vap->va_atime.tv_sec = 0;
1265 			}
1266 			error = vattr_to_sattr(vap, &args.saa_sa);
1267 			/* set vap times back on */
1268 			vap->va_mtime = mtime;
1269 			vap->va_atime = atime;
1270 		}
1271 	} else {
1272 		/* Either do not set times or use the client specified times */
1273 		error = vattr_to_sattr(vap, &args.saa_sa);
1274 	}
1275 	if (error) {
1276 		/* req time field(s) overflow - return immediately */
1277 		return (error);
1278 	}
1279 	args.saa_fh = *VTOFH(vp);
1280 
1281 	va.va_mask = AT_MODE;
1282 	error = nfsgetattr(vp, &va, cr);
1283 	if (error)
1284 		return (error);
1285 	omode = va.va_mode;
1286 
1287 	mi = VTOMI(vp);
1288 
1289 	douprintf = 1;
1290 
1291 	t = gethrtime();
1292 
1293 	error = rfs2call(mi, RFS_SETATTR,
1294 	    xdr_saargs, (caddr_t)&args,
1295 	    xdr_attrstat, (caddr_t)&ns, cr,
1296 	    &douprintf, &ns.ns_status, 0, NULL);
1297 
1298 	/*
1299 	 * Purge the access cache and ACL cache if changing either the
1300 	 * owner of the file, the group owner, or the mode.  These may
1301 	 * change the access permissions of the file, so purge old
1302 	 * information and start over again.
1303 	 */
1304 	if ((mask & (AT_UID | AT_GID | AT_MODE)) && (mi->mi_flags & MI_ACL)) {
1305 		(void) nfs_access_purge_rp(rp);
1306 		if (rp->r_secattr != NULL) {
1307 			mutex_enter(&rp->r_statelock);
1308 			vsp = rp->r_secattr;
1309 			rp->r_secattr = NULL;
1310 			mutex_exit(&rp->r_statelock);
1311 			if (vsp != NULL)
1312 				nfs_acl_free(vsp);
1313 		}
1314 	}
1315 
1316 	if (!error) {
1317 		error = geterrno(ns.ns_status);
1318 		if (!error) {
1319 			/*
1320 			 * If changing the size of the file, invalidate
1321 			 * any local cached data which is no longer part
1322 			 * of the file.  We also possibly invalidate the
1323 			 * last page in the file.  We could use
1324 			 * pvn_vpzero(), but this would mark the page as
1325 			 * modified and require it to be written back to
1326 			 * the server for no particularly good reason.
1327 			 * This way, if we access it, then we bring it
1328 			 * back in.  A read should be cheaper than a
1329 			 * write.
1330 			 */
1331 			if (mask & AT_SIZE) {
1332 				nfs_invalidate_pages(vp,
1333 				    (vap->va_size & PAGEMASK), cr);
1334 			}
1335 			(void) nfs_cache_fattr(vp, &ns.ns_attr, &va, t, cr);
1336 			/*
1337 			 * If NFS_ACL is supported on the server, then the
1338 			 * attributes returned by server may have minimal
1339 			 * permissions sometimes denying access to users having
1340 			 * proper access.  To get the proper attributes, mark
1341 			 * the attributes as expired so that they will be
1342 			 * regotten via the NFS_ACL GETATTR2 procedure.
1343 			 */
1344 			if (mi->mi_flags & MI_ACL) {
1345 				PURGE_ATTRCACHE(vp);
1346 			}
1347 			/*
1348 			 * This next check attempts to deal with NFS
1349 			 * servers which can not handle increasing
1350 			 * the size of the file via setattr.  Most
1351 			 * of these servers do not return an error,
1352 			 * but do not change the size of the file.
1353 			 * Hence, this check and then attempt to set
1354 			 * the file size by writing 1 byte at the
1355 			 * offset of the end of the file that we need.
1356 			 */
1357 			if ((mask & AT_SIZE) &&
1358 			    ns.ns_attr.na_size < (uint32_t)vap->va_size) {
1359 				char zb = '\0';
1360 
1361 				error = nfswrite(vp, &zb,
1362 				    vap->va_size - sizeof (zb),
1363 				    sizeof (zb), cr);
1364 			}
1365 			/*
1366 			 * Some servers will change the mode to clear the setuid
1367 			 * and setgid bits when changing the uid or gid.  The
1368 			 * client needs to compensate appropriately.
1369 			 */
1370 			if (mask & (AT_UID | AT_GID)) {
1371 				int terror;
1372 
1373 				va.va_mask = AT_MODE;
1374 				terror = nfsgetattr(vp, &va, cr);
1375 				if (!terror &&
1376 				    (((mask & AT_MODE) &&
1377 				    va.va_mode != vap->va_mode) ||
1378 				    (!(mask & AT_MODE) &&
1379 				    va.va_mode != omode))) {
1380 					va.va_mask = AT_MODE;
1381 					if (mask & AT_MODE)
1382 						va.va_mode = vap->va_mode;
1383 					else
1384 						va.va_mode = omode;
1385 					(void) nfssetattr(vp, &va, 0, cr);
1386 				}
1387 			}
1388 		} else {
1389 			PURGE_ATTRCACHE(vp);
1390 			PURGE_STALE_FH(error, vp, cr);
1391 		}
1392 	} else {
1393 		PURGE_ATTRCACHE(vp);
1394 	}
1395 
1396 	return (error);
1397 }
1398 
1399 static int
1400 nfs_accessx(void *vp, int mode, cred_t *cr)
1401 {
1402 	ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone);
1403 	return (nfs_access(vp, mode, 0, cr, NULL));
1404 }
1405 
1406 /* ARGSUSED */
1407 static int
1408 nfs_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
1409 {
1410 	struct vattr va;
1411 	int error;
1412 	mntinfo_t *mi;
1413 	int shift = 0;
1414 
1415 	mi = VTOMI(vp);
1416 
1417 	if (nfs_zone() != mi->mi_zone)
1418 		return (EIO);
1419 	if (mi->mi_flags & MI_ACL) {
1420 		error = acl_access2(vp, mode, flags, cr);
1421 		if (mi->mi_flags & MI_ACL)
1422 			return (error);
1423 	}
1424 
1425 	va.va_mask = AT_MODE | AT_UID | AT_GID;
1426 	error = nfsgetattr(vp, &va, cr);
1427 	if (error)
1428 		return (error);
1429 
1430 	/*
1431 	 * Disallow write attempts on read-only
1432 	 * file systems, unless the file is a
1433 	 * device node.
1434 	 */
1435 	if ((mode & VWRITE) && vn_is_readonly(vp) && !IS_DEVVP(vp))
1436 		return (EROFS);
1437 
1438 	/*
1439 	 * Disallow attempts to access mandatory lock files.
1440 	 */
1441 	if ((mode & (VWRITE | VREAD | VEXEC)) &&
1442 	    MANDLOCK(vp, va.va_mode))
1443 		return (EACCES);
1444 
1445 	/*
1446 	 * Access check is based on only
1447 	 * one of owner, group, public.
1448 	 * If not owner, then check group.
1449 	 * If not a member of the group,
1450 	 * then check public access.
1451 	 */
1452 	if (crgetuid(cr) != va.va_uid) {
1453 		shift += 3;
1454 		if (!groupmember(va.va_gid, cr))
1455 			shift += 3;
1456 	}
1457 
1458 	return (secpolicy_vnode_access2(cr, vp, va.va_uid,
1459 	    va.va_mode << shift, mode));
1460 }
1461 
1462 static int nfs_do_symlink_cache = 1;
1463 
1464 /* ARGSUSED */
1465 static int
1466 nfs_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct)
1467 {
1468 	int error;
1469 	struct nfsrdlnres rl;
1470 	rnode_t *rp;
1471 	int douprintf;
1472 	failinfo_t fi;
1473 
1474 	/*
1475 	 * We want to be consistent with UFS semantics so we will return
1476 	 * EINVAL instead of ENXIO. This violates the XNFS spec and
1477 	 * the RFC 1094, which are wrong any way. BUGID 1138002.
1478 	 */
1479 	if (vp->v_type != VLNK)
1480 		return (EINVAL);
1481 
1482 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1483 		return (EIO);
1484 
1485 	rp = VTOR(vp);
1486 	if (nfs_do_symlink_cache && rp->r_symlink.contents != NULL) {
1487 		error = nfs_validate_caches(vp, cr);
1488 		if (error)
1489 			return (error);
1490 		mutex_enter(&rp->r_statelock);
1491 		if (rp->r_symlink.contents != NULL) {
1492 			error = uiomove(rp->r_symlink.contents,
1493 			    rp->r_symlink.len, UIO_READ, uiop);
1494 			mutex_exit(&rp->r_statelock);
1495 			return (error);
1496 		}
1497 		mutex_exit(&rp->r_statelock);
1498 	}
1499 
1500 
1501 	rl.rl_data = kmem_alloc(NFS_MAXPATHLEN, KM_SLEEP);
1502 
1503 	fi.vp = vp;
1504 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1505 	fi.copyproc = nfscopyfh;
1506 	fi.lookupproc = nfslookup;
1507 	fi.xattrdirproc = acl_getxattrdir2;
1508 
1509 	douprintf = 1;
1510 
1511 	error = rfs2call(VTOMI(vp), RFS_READLINK,
1512 	    xdr_readlink, (caddr_t)VTOFH(vp),
1513 	    xdr_rdlnres, (caddr_t)&rl, cr,
1514 	    &douprintf, &rl.rl_status, 0, &fi);
1515 
1516 	if (error) {
1517 
1518 		kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1519 		return (error);
1520 	}
1521 
1522 	error = geterrno(rl.rl_status);
1523 	if (!error) {
1524 		error = uiomove(rl.rl_data, (int)rl.rl_count, UIO_READ, uiop);
1525 		if (nfs_do_symlink_cache && rp->r_symlink.contents == NULL) {
1526 			mutex_enter(&rp->r_statelock);
1527 			if (rp->r_symlink.contents == NULL) {
1528 				rp->r_symlink.contents = rl.rl_data;
1529 				rp->r_symlink.len = (int)rl.rl_count;
1530 				rp->r_symlink.size = NFS_MAXPATHLEN;
1531 				mutex_exit(&rp->r_statelock);
1532 			} else {
1533 				mutex_exit(&rp->r_statelock);
1534 
1535 				kmem_free((void *)rl.rl_data,
1536 				    NFS_MAXPATHLEN);
1537 			}
1538 		} else {
1539 
1540 			kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1541 		}
1542 	} else {
1543 		PURGE_STALE_FH(error, vp, cr);
1544 
1545 		kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1546 	}
1547 
1548 	/*
1549 	 * Conform to UFS semantics (see comment above)
1550 	 */
1551 	return (error == ENXIO ? EINVAL : error);
1552 }
1553 
1554 /*
1555  * Flush local dirty pages to stable storage on the server.
1556  *
1557  * If FNODSYNC is specified, then there is nothing to do because
1558  * metadata changes are not cached on the client before being
1559  * sent to the server.
1560  */
1561 /* ARGSUSED */
1562 static int
1563 nfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
1564 {
1565 	int error;
1566 
1567 	if ((syncflag & FNODSYNC) || IS_SWAPVP(vp))
1568 		return (0);
1569 
1570 	if (nfs_zone() != VTOMI(vp)->mi_zone)
1571 		return (EIO);
1572 
1573 	error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1574 	if (!error)
1575 		error = VTOR(vp)->r_error;
1576 	return (error);
1577 }
1578 
1579 
1580 /*
1581  * Weirdness: if the file was removed or the target of a rename
1582  * operation while it was open, it got renamed instead.  Here we
1583  * remove the renamed file.
1584  */
1585 /* ARGSUSED */
1586 static void
1587 nfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1588 {
1589 	rnode_t *rp;
1590 
1591 	ASSERT(vp != DNLC_NO_VNODE);
1592 
1593 	/*
1594 	 * If this is coming from the wrong zone, we let someone in the right
1595 	 * zone take care of it asynchronously.  We can get here due to
1596 	 * VN_RELE() being called from pageout() or fsflush().  This call may
1597 	 * potentially turn into an expensive no-op if, for instance, v_count
1598 	 * gets incremented in the meantime, but it's still correct.
1599 	 */
1600 	if (nfs_zone() != VTOMI(vp)->mi_zone) {
1601 		nfs_async_inactive(vp, cr, nfs_inactive);
1602 		return;
1603 	}
1604 
1605 	rp = VTOR(vp);
1606 redo:
1607 	if (rp->r_unldvp != NULL) {
1608 		/*
1609 		 * Save the vnode pointer for the directory where the
1610 		 * unlinked-open file got renamed, then set it to NULL
1611 		 * to prevent another thread from getting here before
1612 		 * we're done with the remove.  While we have the
1613 		 * statelock, make local copies of the pertinent rnode
1614 		 * fields.  If we weren't to do this in an atomic way, the
1615 		 * the unl* fields could become inconsistent with respect
1616 		 * to each other due to a race condition between this
1617 		 * code and nfs_remove().  See bug report 1034328.
1618 		 */
1619 		mutex_enter(&rp->r_statelock);
1620 		if (rp->r_unldvp != NULL) {
1621 			vnode_t *unldvp;
1622 			char *unlname;
1623 			cred_t *unlcred;
1624 			struct nfsdiropargs da;
1625 			enum nfsstat status;
1626 			int douprintf;
1627 			int error;
1628 
1629 			unldvp = rp->r_unldvp;
1630 			rp->r_unldvp = NULL;
1631 			unlname = rp->r_unlname;
1632 			rp->r_unlname = NULL;
1633 			unlcred = rp->r_unlcred;
1634 			rp->r_unlcred = NULL;
1635 			mutex_exit(&rp->r_statelock);
1636 
1637 			/*
1638 			 * If there are any dirty pages left, then flush
1639 			 * them.  This is unfortunate because they just
1640 			 * may get thrown away during the remove operation,
1641 			 * but we have to do this for correctness.
1642 			 */
1643 			if (vn_has_cached_data(vp) &&
1644 			    ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
1645 				ASSERT(vp->v_type != VCHR);
1646 				error = nfs_putpage(vp, (offset_t)0, 0, 0,
1647 				    cr, ct);
1648 				if (error) {
1649 					mutex_enter(&rp->r_statelock);
1650 					if (!rp->r_error)
1651 						rp->r_error = error;
1652 					mutex_exit(&rp->r_statelock);
1653 				}
1654 			}
1655 
1656 			/*
1657 			 * Do the remove operation on the renamed file
1658 			 */
1659 			setdiropargs(&da, unlname, unldvp);
1660 
1661 			douprintf = 1;
1662 
1663 			(void) rfs2call(VTOMI(unldvp), RFS_REMOVE,
1664 			    xdr_diropargs, (caddr_t)&da,
1665 			    xdr_enum, (caddr_t)&status, unlcred,
1666 			    &douprintf, &status, 0, NULL);
1667 
1668 			if (HAVE_RDDIR_CACHE(VTOR(unldvp)))
1669 				nfs_purge_rddir_cache(unldvp);
1670 			PURGE_ATTRCACHE(unldvp);
1671 
1672 			/*
1673 			 * Release stuff held for the remove
1674 			 */
1675 			VN_RELE(unldvp);
1676 			kmem_free(unlname, MAXNAMELEN);
1677 			crfree(unlcred);
1678 			goto redo;
1679 		}
1680 		mutex_exit(&rp->r_statelock);
1681 	}
1682 
1683 	rp_addfree(rp, cr);
1684 }
1685 
1686 /*
1687  * Remote file system operations having to do with directory manipulation.
1688  */
1689 
1690 /* ARGSUSED */
1691 static int
1692 nfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1693 	int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1694 	int *direntflags, pathname_t *realpnp)
1695 {
1696 	int error;
1697 	vnode_t *vp;
1698 	vnode_t *avp = NULL;
1699 	rnode_t *drp;
1700 
1701 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
1702 		return (EPERM);
1703 
1704 	drp = VTOR(dvp);
1705 
1706 	/*
1707 	 * Are we looking up extended attributes?  If so, "dvp" is
1708 	 * the file or directory for which we want attributes, and
1709 	 * we need a lookup of the hidden attribute directory
1710 	 * before we lookup the rest of the path.
1711 	 */
1712 	if (flags & LOOKUP_XATTR) {
1713 		bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0);
1714 		mntinfo_t *mi;
1715 
1716 		mi = VTOMI(dvp);
1717 		if (!(mi->mi_flags & MI_EXTATTR))
1718 			return (EINVAL);
1719 
1720 		if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp)))
1721 			return (EINTR);
1722 
1723 		(void) nfslookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr);
1724 		if (avp == NULL)
1725 			error = acl_getxattrdir2(dvp, &avp, cflag, cr, 0);
1726 		else
1727 			error = 0;
1728 
1729 		nfs_rw_exit(&drp->r_rwlock);
1730 
1731 		if (error) {
1732 			if (mi->mi_flags & MI_EXTATTR)
1733 				return (error);
1734 			return (EINVAL);
1735 		}
1736 		dvp = avp;
1737 		drp = VTOR(dvp);
1738 	}
1739 
1740 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) {
1741 		error = EINTR;
1742 		goto out;
1743 	}
1744 
1745 	error = nfslookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0);
1746 
1747 	nfs_rw_exit(&drp->r_rwlock);
1748 
1749 	/*
1750 	 * If vnode is a device, create special vnode.
1751 	 */
1752 	if (!error && IS_DEVVP(*vpp)) {
1753 		vp = *vpp;
1754 		*vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
1755 		VN_RELE(vp);
1756 	}
1757 
1758 out:
1759 	if (avp != NULL)
1760 		VN_RELE(avp);
1761 
1762 	return (error);
1763 }
1764 
1765 static int nfs_lookup_neg_cache = 1;
1766 
1767 #ifdef DEBUG
1768 static int nfs_lookup_dnlc_hits = 0;
1769 static int nfs_lookup_dnlc_misses = 0;
1770 static int nfs_lookup_dnlc_neg_hits = 0;
1771 static int nfs_lookup_dnlc_disappears = 0;
1772 static int nfs_lookup_dnlc_lookups = 0;
1773 #endif
1774 
1775 /* ARGSUSED */
1776 int
1777 nfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1778 	int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags)
1779 {
1780 	int error;
1781 
1782 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1783 
1784 	/*
1785 	 * If lookup is for "", just return dvp.  Don't need
1786 	 * to send it over the wire, look it up in the dnlc,
1787 	 * or perform any access checks.
1788 	 */
1789 	if (*nm == '\0') {
1790 		VN_HOLD(dvp);
1791 		*vpp = dvp;
1792 		return (0);
1793 	}
1794 
1795 	/*
1796 	 * Can't do lookups in non-directories.
1797 	 */
1798 	if (dvp->v_type != VDIR)
1799 		return (ENOTDIR);
1800 
1801 	/*
1802 	 * If we're called with RFSCALL_SOFT, it's important that
1803 	 * the only rfscall is one we make directly; if we permit
1804 	 * an access call because we're looking up "." or validating
1805 	 * a dnlc hit, we'll deadlock because that rfscall will not
1806 	 * have the RFSCALL_SOFT set.
1807 	 */
1808 	if (rfscall_flags & RFSCALL_SOFT)
1809 		goto callit;
1810 
1811 	/*
1812 	 * If lookup is for ".", just return dvp.  Don't need
1813 	 * to send it over the wire or look it up in the dnlc,
1814 	 * just need to check access.
1815 	 */
1816 	if (strcmp(nm, ".") == 0) {
1817 		error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1818 		if (error)
1819 			return (error);
1820 		VN_HOLD(dvp);
1821 		*vpp = dvp;
1822 		return (0);
1823 	}
1824 
1825 	/*
1826 	 * Lookup this name in the DNLC.  If there was a valid entry,
1827 	 * then return the results of the lookup.
1828 	 */
1829 	error = nfslookup_dnlc(dvp, nm, vpp, cr);
1830 	if (error || *vpp != NULL)
1831 		return (error);
1832 
1833 callit:
1834 	error = nfslookup_otw(dvp, nm, vpp, cr, rfscall_flags);
1835 
1836 	return (error);
1837 }
1838 
1839 static int
1840 nfslookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr)
1841 {
1842 	int error;
1843 	vnode_t *vp;
1844 
1845 	ASSERT(*nm != '\0');
1846 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1847 
1848 	/*
1849 	 * Lookup this name in the DNLC.  If successful, then validate
1850 	 * the caches and then recheck the DNLC.  The DNLC is rechecked
1851 	 * just in case this entry got invalidated during the call
1852 	 * to nfs_validate_caches.
1853 	 *
1854 	 * An assumption is being made that it is safe to say that a
1855 	 * file exists which may not on the server.  Any operations to
1856 	 * the server will fail with ESTALE.
1857 	 */
1858 #ifdef DEBUG
1859 	nfs_lookup_dnlc_lookups++;
1860 #endif
1861 	vp = dnlc_lookup(dvp, nm);
1862 	if (vp != NULL) {
1863 		VN_RELE(vp);
1864 		if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) {
1865 			PURGE_ATTRCACHE(dvp);
1866 		}
1867 		error = nfs_validate_caches(dvp, cr);
1868 		if (error)
1869 			return (error);
1870 		vp = dnlc_lookup(dvp, nm);
1871 		if (vp != NULL) {
1872 			error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1873 			if (error) {
1874 				VN_RELE(vp);
1875 				return (error);
1876 			}
1877 			if (vp == DNLC_NO_VNODE) {
1878 				VN_RELE(vp);
1879 #ifdef DEBUG
1880 				nfs_lookup_dnlc_neg_hits++;
1881 #endif
1882 				return (ENOENT);
1883 			}
1884 			*vpp = vp;
1885 #ifdef DEBUG
1886 			nfs_lookup_dnlc_hits++;
1887 #endif
1888 			return (0);
1889 		}
1890 #ifdef DEBUG
1891 		nfs_lookup_dnlc_disappears++;
1892 #endif
1893 	}
1894 #ifdef DEBUG
1895 	else
1896 		nfs_lookup_dnlc_misses++;
1897 #endif
1898 
1899 	*vpp = NULL;
1900 
1901 	return (0);
1902 }
1903 
1904 static int
1905 nfslookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr,
1906 	int rfscall_flags)
1907 {
1908 	int error;
1909 	struct nfsdiropargs da;
1910 	struct nfsdiropres dr;
1911 	int douprintf;
1912 	failinfo_t fi;
1913 	hrtime_t t;
1914 
1915 	ASSERT(*nm != '\0');
1916 	ASSERT(dvp->v_type == VDIR);
1917 	ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1918 
1919 	setdiropargs(&da, nm, dvp);
1920 
1921 	fi.vp = dvp;
1922 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1923 	fi.copyproc = nfscopyfh;
1924 	fi.lookupproc = nfslookup;
1925 	fi.xattrdirproc = acl_getxattrdir2;
1926 
1927 	douprintf = 1;
1928 
1929 	t = gethrtime();
1930 
1931 	error = rfs2call(VTOMI(dvp), RFS_LOOKUP,
1932 	    xdr_diropargs, (caddr_t)&da,
1933 	    xdr_diropres, (caddr_t)&dr, cr,
1934 	    &douprintf, &dr.dr_status, rfscall_flags, &fi);
1935 
1936 	if (!error) {
1937 		error = geterrno(dr.dr_status);
1938 		if (!error) {
1939 			*vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
1940 			    dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm);
1941 			/*
1942 			 * If NFS_ACL is supported on the server, then the
1943 			 * attributes returned by server may have minimal
1944 			 * permissions sometimes denying access to users having
1945 			 * proper access.  To get the proper attributes, mark
1946 			 * the attributes as expired so that they will be
1947 			 * regotten via the NFS_ACL GETATTR2 procedure.
1948 			 */
1949 			if (VTOMI(*vpp)->mi_flags & MI_ACL) {
1950 				PURGE_ATTRCACHE(*vpp);
1951 			}
1952 			if (!(rfscall_flags & RFSCALL_SOFT))
1953 				dnlc_update(dvp, nm, *vpp);
1954 		} else {
1955 			PURGE_STALE_FH(error, dvp, cr);
1956 			if (error == ENOENT && nfs_lookup_neg_cache)
1957 				dnlc_enter(dvp, nm, DNLC_NO_VNODE);
1958 		}
1959 	}
1960 
1961 	return (error);
1962 }
1963 
1964 /* ARGSUSED */
1965 static int
1966 nfs_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
1967 	int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct,
1968 	vsecattr_t *vsecp)
1969 {
1970 	int error;
1971 	struct nfscreatargs args;
1972 	struct nfsdiropres dr;
1973 	int douprintf;
1974 	vnode_t *vp;
1975 	rnode_t *rp;
1976 	struct vattr vattr;
1977 	rnode_t *drp;
1978 	vnode_t *tempvp;
1979 	hrtime_t t;
1980 
1981 	drp = VTOR(dvp);
1982 
1983 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
1984 		return (EPERM);
1985 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
1986 		return (EINTR);
1987 
1988 	/*
1989 	 * We make a copy of the attributes because the caller does not
1990 	 * expect us to change what va points to.
1991 	 */
1992 	vattr = *va;
1993 
1994 	/*
1995 	 * If the pathname is "", just use dvp.  Don't need
1996 	 * to send it over the wire, look it up in the dnlc,
1997 	 * or perform any access checks.
1998 	 */
1999 	if (*nm == '\0') {
2000 		error = 0;
2001 		VN_HOLD(dvp);
2002 		vp = dvp;
2003 	/*
2004 	 * If the pathname is ".", just use dvp.  Don't need
2005 	 * to send it over the wire or look it up in the dnlc,
2006 	 * just need to check access.
2007 	 */
2008 	} else if (strcmp(nm, ".") == 0) {
2009 		error = nfs_access(dvp, VEXEC, 0, cr, ct);
2010 		if (error) {
2011 			nfs_rw_exit(&drp->r_rwlock);
2012 			return (error);
2013 		}
2014 		VN_HOLD(dvp);
2015 		vp = dvp;
2016 	/*
2017 	 * We need to go over the wire, just to be sure whether the
2018 	 * file exists or not.  Using the DNLC can be dangerous in
2019 	 * this case when making a decision regarding existence.
2020 	 */
2021 	} else {
2022 		error = nfslookup_otw(dvp, nm, &vp, cr, 0);
2023 	}
2024 	if (!error) {
2025 		if (exclusive == EXCL)
2026 			error = EEXIST;
2027 		else if (vp->v_type == VDIR && (mode & VWRITE))
2028 			error = EISDIR;
2029 		else {
2030 			/*
2031 			 * If vnode is a device, create special vnode.
2032 			 */
2033 			if (IS_DEVVP(vp)) {
2034 				tempvp = vp;
2035 				vp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2036 				VN_RELE(tempvp);
2037 			}
2038 			if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) {
2039 				if ((vattr.va_mask & AT_SIZE) &&
2040 				    vp->v_type == VREG) {
2041 					vattr.va_mask = AT_SIZE;
2042 					error = nfssetattr(vp, &vattr, 0, cr);
2043 
2044 					if (!error) {
2045 						/*
2046 						 * Existing file was truncated;
2047 						 * emit a create event.
2048 						 */
2049 						vnevent_create(vp, ct);
2050 					}
2051 				}
2052 			}
2053 		}
2054 		nfs_rw_exit(&drp->r_rwlock);
2055 		if (error) {
2056 			VN_RELE(vp);
2057 		} else {
2058 			*vpp = vp;
2059 		}
2060 		return (error);
2061 	}
2062 
2063 	ASSERT(vattr.va_mask & AT_TYPE);
2064 	if (vattr.va_type == VREG) {
2065 		ASSERT(vattr.va_mask & AT_MODE);
2066 		if (MANDMODE(vattr.va_mode)) {
2067 			nfs_rw_exit(&drp->r_rwlock);
2068 			return (EACCES);
2069 		}
2070 	}
2071 
2072 	dnlc_remove(dvp, nm);
2073 
2074 	setdiropargs(&args.ca_da, nm, dvp);
2075 
2076 	/*
2077 	 * Decide what the group-id of the created file should be.
2078 	 * Set it in attribute list as advisory...then do a setattr
2079 	 * if the server didn't get it right the first time.
2080 	 */
2081 	error = setdirgid(dvp, &vattr.va_gid, cr);
2082 	if (error) {
2083 		nfs_rw_exit(&drp->r_rwlock);
2084 		return (error);
2085 	}
2086 	vattr.va_mask |= AT_GID;
2087 
2088 	/*
2089 	 * This is a completely gross hack to make mknod
2090 	 * work over the wire until we can wack the protocol
2091 	 */
2092 #define	IFCHR		0020000		/* character special */
2093 #define	IFBLK		0060000		/* block special */
2094 #define	IFSOCK		0140000		/* socket */
2095 
2096 	/*
2097 	 * dev_t is uint_t in 5.x and short in 4.x. Both 4.x
2098 	 * supports 8 bit majors. 5.x supports 14 bit majors. 5.x supports 18
2099 	 * bits in the minor number where 4.x supports 8 bits.  If the 5.x
2100 	 * minor/major numbers <= 8 bits long, compress the device
2101 	 * number before sending it. Otherwise, the 4.x server will not
2102 	 * create the device with the correct device number and nothing can be
2103 	 * done about this.
2104 	 */
2105 	if (vattr.va_type == VCHR || vattr.va_type == VBLK) {
2106 		dev_t d = vattr.va_rdev;
2107 		dev32_t dev32;
2108 
2109 		if (vattr.va_type == VCHR)
2110 			vattr.va_mode |= IFCHR;
2111 		else
2112 			vattr.va_mode |= IFBLK;
2113 
2114 		(void) cmpldev(&dev32, d);
2115 		if (dev32 & ~((SO4_MAXMAJ << L_BITSMINOR32) | SO4_MAXMIN))
2116 			vattr.va_size = (u_offset_t)dev32;
2117 		else
2118 			vattr.va_size = (u_offset_t)nfsv2_cmpdev(d);
2119 
2120 		vattr.va_mask |= AT_MODE|AT_SIZE;
2121 	} else if (vattr.va_type == VFIFO) {
2122 		vattr.va_mode |= IFCHR;		/* xtra kludge for namedpipe */
2123 		vattr.va_size = (u_offset_t)NFS_FIFO_DEV;	/* blech */
2124 		vattr.va_mask |= AT_MODE|AT_SIZE;
2125 	} else if (vattr.va_type == VSOCK) {
2126 		vattr.va_mode |= IFSOCK;
2127 		/*
2128 		 * To avoid triggering bugs in the servers set AT_SIZE
2129 		 * (all other RFS_CREATE calls set this).
2130 		 */
2131 		vattr.va_size = 0;
2132 		vattr.va_mask |= AT_MODE|AT_SIZE;
2133 	}
2134 
2135 	args.ca_sa = &args.ca_sa_buf;
2136 	error = vattr_to_sattr(&vattr, args.ca_sa);
2137 	if (error) {
2138 		/* req time field(s) overflow - return immediately */
2139 		nfs_rw_exit(&drp->r_rwlock);
2140 		return (error);
2141 	}
2142 
2143 	douprintf = 1;
2144 
2145 	t = gethrtime();
2146 
2147 	error = rfs2call(VTOMI(dvp), RFS_CREATE,
2148 	    xdr_creatargs, (caddr_t)&args,
2149 	    xdr_diropres, (caddr_t)&dr, cr,
2150 	    &douprintf, &dr.dr_status, 0, NULL);
2151 
2152 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2153 
2154 	if (!error) {
2155 		error = geterrno(dr.dr_status);
2156 		if (!error) {
2157 			if (HAVE_RDDIR_CACHE(drp))
2158 				nfs_purge_rddir_cache(dvp);
2159 			vp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2160 			    dvp->v_vfsp, t, cr, NULL, NULL);
2161 			/*
2162 			 * If NFS_ACL is supported on the server, then the
2163 			 * attributes returned by server may have minimal
2164 			 * permissions sometimes denying access to users having
2165 			 * proper access.  To get the proper attributes, mark
2166 			 * the attributes as expired so that they will be
2167 			 * regotten via the NFS_ACL GETATTR2 procedure.
2168 			 */
2169 			if (VTOMI(vp)->mi_flags & MI_ACL) {
2170 				PURGE_ATTRCACHE(vp);
2171 			}
2172 			dnlc_update(dvp, nm, vp);
2173 			rp = VTOR(vp);
2174 			if (vattr.va_size == 0) {
2175 				mutex_enter(&rp->r_statelock);
2176 				rp->r_size = 0;
2177 				mutex_exit(&rp->r_statelock);
2178 				if (vn_has_cached_data(vp)) {
2179 					ASSERT(vp->v_type != VCHR);
2180 					nfs_invalidate_pages(vp,
2181 					    (u_offset_t)0, cr);
2182 				}
2183 			}
2184 
2185 			/*
2186 			 * Make sure the gid was set correctly.
2187 			 * If not, try to set it (but don't lose
2188 			 * any sleep over it).
2189 			 */
2190 			if (vattr.va_gid != rp->r_attr.va_gid) {
2191 				vattr.va_mask = AT_GID;
2192 				(void) nfssetattr(vp, &vattr, 0, cr);
2193 			}
2194 
2195 			/*
2196 			 * If vnode is a device create special vnode
2197 			 */
2198 			if (IS_DEVVP(vp)) {
2199 				*vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2200 				VN_RELE(vp);
2201 			} else
2202 				*vpp = vp;
2203 		} else {
2204 			PURGE_STALE_FH(error, dvp, cr);
2205 		}
2206 	}
2207 
2208 	nfs_rw_exit(&drp->r_rwlock);
2209 
2210 	return (error);
2211 }
2212 
2213 /*
2214  * Weirdness: if the vnode to be removed is open
2215  * we rename it instead of removing it and nfs_inactive
2216  * will remove the new name.
2217  */
2218 /* ARGSUSED */
2219 static int
2220 nfs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags)
2221 {
2222 	int error;
2223 	struct nfsdiropargs da;
2224 	enum nfsstat status;
2225 	vnode_t *vp;
2226 	char *tmpname;
2227 	int douprintf;
2228 	rnode_t *rp;
2229 	rnode_t *drp;
2230 
2231 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2232 		return (EPERM);
2233 	drp = VTOR(dvp);
2234 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2235 		return (EINTR);
2236 
2237 	error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2238 	if (error) {
2239 		nfs_rw_exit(&drp->r_rwlock);
2240 		return (error);
2241 	}
2242 
2243 	if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) {
2244 		VN_RELE(vp);
2245 		nfs_rw_exit(&drp->r_rwlock);
2246 		return (EPERM);
2247 	}
2248 
2249 	/*
2250 	 * First just remove the entry from the name cache, as it
2251 	 * is most likely the only entry for this vp.
2252 	 */
2253 	dnlc_remove(dvp, nm);
2254 
2255 	/*
2256 	 * If the file has a v_count > 1 then there may be more than one
2257 	 * entry in the name cache due multiple links or an open file,
2258 	 * but we don't have the real reference count so flush all
2259 	 * possible entries.
2260 	 */
2261 	if (vp->v_count > 1)
2262 		dnlc_purge_vp(vp);
2263 
2264 	/*
2265 	 * Now we have the real reference count on the vnode
2266 	 */
2267 	rp = VTOR(vp);
2268 	mutex_enter(&rp->r_statelock);
2269 	if (vp->v_count > 1 &&
2270 	    (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) {
2271 		mutex_exit(&rp->r_statelock);
2272 		tmpname = newname();
2273 		error = nfsrename(dvp, nm, dvp, tmpname, cr, ct);
2274 		if (error)
2275 			kmem_free(tmpname, MAXNAMELEN);
2276 		else {
2277 			mutex_enter(&rp->r_statelock);
2278 			if (rp->r_unldvp == NULL) {
2279 				VN_HOLD(dvp);
2280 				rp->r_unldvp = dvp;
2281 				if (rp->r_unlcred != NULL)
2282 					crfree(rp->r_unlcred);
2283 				crhold(cr);
2284 				rp->r_unlcred = cr;
2285 				rp->r_unlname = tmpname;
2286 			} else {
2287 				kmem_free(rp->r_unlname, MAXNAMELEN);
2288 				rp->r_unlname = tmpname;
2289 			}
2290 			mutex_exit(&rp->r_statelock);
2291 		}
2292 	} else {
2293 		mutex_exit(&rp->r_statelock);
2294 		/*
2295 		 * We need to flush any dirty pages which happen to
2296 		 * be hanging around before removing the file.  This
2297 		 * shouldn't happen very often and mostly on file
2298 		 * systems mounted "nocto".
2299 		 */
2300 		if (vn_has_cached_data(vp) &&
2301 		    ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
2302 			error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
2303 			if (error && (error == ENOSPC || error == EDQUOT)) {
2304 				mutex_enter(&rp->r_statelock);
2305 				if (!rp->r_error)
2306 					rp->r_error = error;
2307 				mutex_exit(&rp->r_statelock);
2308 			}
2309 		}
2310 
2311 		setdiropargs(&da, nm, dvp);
2312 
2313 		douprintf = 1;
2314 
2315 		error = rfs2call(VTOMI(dvp), RFS_REMOVE,
2316 		    xdr_diropargs, (caddr_t)&da,
2317 		    xdr_enum, (caddr_t)&status, cr,
2318 		    &douprintf, &status, 0, NULL);
2319 
2320 		/*
2321 		 * The xattr dir may be gone after last attr is removed,
2322 		 * so flush it from dnlc.
2323 		 */
2324 		if (dvp->v_flag & V_XATTRDIR)
2325 			dnlc_purge_vp(dvp);
2326 
2327 		PURGE_ATTRCACHE(dvp);	/* mod time changed */
2328 		PURGE_ATTRCACHE(vp);	/* link count changed */
2329 
2330 		if (!error) {
2331 			error = geterrno(status);
2332 			if (!error) {
2333 				if (HAVE_RDDIR_CACHE(drp))
2334 					nfs_purge_rddir_cache(dvp);
2335 			} else {
2336 				PURGE_STALE_FH(error, dvp, cr);
2337 			}
2338 		}
2339 	}
2340 
2341 	if (error == 0) {
2342 		vnevent_remove(vp, dvp, nm, ct);
2343 	}
2344 	VN_RELE(vp);
2345 
2346 	nfs_rw_exit(&drp->r_rwlock);
2347 
2348 	return (error);
2349 }
2350 
2351 /* ARGSUSED */
2352 static int
2353 nfs_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr,
2354 	caller_context_t *ct, int flags)
2355 {
2356 	int error;
2357 	struct nfslinkargs args;
2358 	enum nfsstat status;
2359 	vnode_t *realvp;
2360 	int douprintf;
2361 	rnode_t *tdrp;
2362 
2363 	if (nfs_zone() != VTOMI(tdvp)->mi_zone)
2364 		return (EPERM);
2365 	if (VOP_REALVP(svp, &realvp, ct) == 0)
2366 		svp = realvp;
2367 
2368 	args.la_from = VTOFH(svp);
2369 	setdiropargs(&args.la_to, tnm, tdvp);
2370 
2371 	tdrp = VTOR(tdvp);
2372 	if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp)))
2373 		return (EINTR);
2374 
2375 	dnlc_remove(tdvp, tnm);
2376 
2377 	douprintf = 1;
2378 
2379 	error = rfs2call(VTOMI(svp), RFS_LINK,
2380 	    xdr_linkargs, (caddr_t)&args,
2381 	    xdr_enum, (caddr_t)&status, cr,
2382 	    &douprintf, &status, 0, NULL);
2383 
2384 	PURGE_ATTRCACHE(tdvp);	/* mod time changed */
2385 	PURGE_ATTRCACHE(svp);	/* link count changed */
2386 
2387 	if (!error) {
2388 		error = geterrno(status);
2389 		if (!error) {
2390 			if (HAVE_RDDIR_CACHE(tdrp))
2391 				nfs_purge_rddir_cache(tdvp);
2392 		}
2393 	}
2394 
2395 	nfs_rw_exit(&tdrp->r_rwlock);
2396 
2397 	if (!error) {
2398 		/*
2399 		 * Notify the source file of this link operation.
2400 		 */
2401 		vnevent_link(svp, ct);
2402 	}
2403 	return (error);
2404 }
2405 
2406 /* ARGSUSED */
2407 static int
2408 nfs_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2409 	caller_context_t *ct, int flags)
2410 {
2411 	vnode_t *realvp;
2412 
2413 	if (nfs_zone() != VTOMI(odvp)->mi_zone)
2414 		return (EPERM);
2415 	if (VOP_REALVP(ndvp, &realvp, ct) == 0)
2416 		ndvp = realvp;
2417 
2418 	return (nfsrename(odvp, onm, ndvp, nnm, cr, ct));
2419 }
2420 
2421 /*
2422  * nfsrename does the real work of renaming in NFS Version 2.
2423  */
2424 static int
2425 nfsrename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2426     caller_context_t *ct)
2427 {
2428 	int error;
2429 	enum nfsstat status;
2430 	struct nfsrnmargs args;
2431 	int douprintf;
2432 	vnode_t *nvp = NULL;
2433 	vnode_t *ovp = NULL;
2434 	char *tmpname;
2435 	rnode_t *rp;
2436 	rnode_t *odrp;
2437 	rnode_t *ndrp;
2438 
2439 	ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone);
2440 	if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 ||
2441 	    strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0)
2442 		return (EINVAL);
2443 
2444 	odrp = VTOR(odvp);
2445 	ndrp = VTOR(ndvp);
2446 	if ((intptr_t)odrp < (intptr_t)ndrp) {
2447 		if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp)))
2448 			return (EINTR);
2449 		if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) {
2450 			nfs_rw_exit(&odrp->r_rwlock);
2451 			return (EINTR);
2452 		}
2453 	} else {
2454 		if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp)))
2455 			return (EINTR);
2456 		if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) {
2457 			nfs_rw_exit(&ndrp->r_rwlock);
2458 			return (EINTR);
2459 		}
2460 	}
2461 
2462 	/*
2463 	 * Lookup the target file.  If it exists, it needs to be
2464 	 * checked to see whether it is a mount point and whether
2465 	 * it is active (open).
2466 	 */
2467 	error = nfslookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0);
2468 	if (!error) {
2469 		/*
2470 		 * If this file has been mounted on, then just
2471 		 * return busy because renaming to it would remove
2472 		 * the mounted file system from the name space.
2473 		 */
2474 		if (vn_mountedvfs(nvp) != NULL) {
2475 			VN_RELE(nvp);
2476 			nfs_rw_exit(&odrp->r_rwlock);
2477 			nfs_rw_exit(&ndrp->r_rwlock);
2478 			return (EBUSY);
2479 		}
2480 
2481 		/*
2482 		 * Purge the name cache of all references to this vnode
2483 		 * so that we can check the reference count to infer
2484 		 * whether it is active or not.
2485 		 */
2486 		/*
2487 		 * First just remove the entry from the name cache, as it
2488 		 * is most likely the only entry for this vp.
2489 		 */
2490 		dnlc_remove(ndvp, nnm);
2491 		/*
2492 		 * If the file has a v_count > 1 then there may be more
2493 		 * than one entry in the name cache due multiple links
2494 		 * or an open file, but we don't have the real reference
2495 		 * count so flush all possible entries.
2496 		 */
2497 		if (nvp->v_count > 1)
2498 			dnlc_purge_vp(nvp);
2499 
2500 		/*
2501 		 * If the vnode is active and is not a directory,
2502 		 * arrange to rename it to a
2503 		 * temporary file so that it will continue to be
2504 		 * accessible.  This implements the "unlink-open-file"
2505 		 * semantics for the target of a rename operation.
2506 		 * Before doing this though, make sure that the
2507 		 * source and target files are not already the same.
2508 		 */
2509 		if (nvp->v_count > 1 && nvp->v_type != VDIR) {
2510 			/*
2511 			 * Lookup the source name.
2512 			 */
2513 			error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL,
2514 			    cr, 0);
2515 
2516 			/*
2517 			 * The source name *should* already exist.
2518 			 */
2519 			if (error) {
2520 				VN_RELE(nvp);
2521 				nfs_rw_exit(&odrp->r_rwlock);
2522 				nfs_rw_exit(&ndrp->r_rwlock);
2523 				return (error);
2524 			}
2525 
2526 			/*
2527 			 * Compare the two vnodes.  If they are the same,
2528 			 * just release all held vnodes and return success.
2529 			 */
2530 			if (ovp == nvp) {
2531 				VN_RELE(ovp);
2532 				VN_RELE(nvp);
2533 				nfs_rw_exit(&odrp->r_rwlock);
2534 				nfs_rw_exit(&ndrp->r_rwlock);
2535 				return (0);
2536 			}
2537 
2538 			/*
2539 			 * Can't mix and match directories and non-
2540 			 * directories in rename operations.  We already
2541 			 * know that the target is not a directory.  If
2542 			 * the source is a directory, return an error.
2543 			 */
2544 			if (ovp->v_type == VDIR) {
2545 				VN_RELE(ovp);
2546 				VN_RELE(nvp);
2547 				nfs_rw_exit(&odrp->r_rwlock);
2548 				nfs_rw_exit(&ndrp->r_rwlock);
2549 				return (ENOTDIR);
2550 			}
2551 
2552 			/*
2553 			 * The target file exists, is not the same as
2554 			 * the source file, and is active.  Link it
2555 			 * to a temporary filename to avoid having
2556 			 * the server removing the file completely.
2557 			 */
2558 			tmpname = newname();
2559 			error = nfs_link(ndvp, nvp, tmpname, cr, NULL, 0);
2560 			if (error == EOPNOTSUPP) {
2561 				error = nfs_rename(ndvp, nnm, ndvp, tmpname,
2562 				    cr, NULL, 0);
2563 			}
2564 			if (error) {
2565 				kmem_free(tmpname, MAXNAMELEN);
2566 				VN_RELE(ovp);
2567 				VN_RELE(nvp);
2568 				nfs_rw_exit(&odrp->r_rwlock);
2569 				nfs_rw_exit(&ndrp->r_rwlock);
2570 				return (error);
2571 			}
2572 			rp = VTOR(nvp);
2573 			mutex_enter(&rp->r_statelock);
2574 			if (rp->r_unldvp == NULL) {
2575 				VN_HOLD(ndvp);
2576 				rp->r_unldvp = ndvp;
2577 				if (rp->r_unlcred != NULL)
2578 					crfree(rp->r_unlcred);
2579 				crhold(cr);
2580 				rp->r_unlcred = cr;
2581 				rp->r_unlname = tmpname;
2582 			} else {
2583 				kmem_free(rp->r_unlname, MAXNAMELEN);
2584 				rp->r_unlname = tmpname;
2585 			}
2586 			mutex_exit(&rp->r_statelock);
2587 		}
2588 	}
2589 
2590 	if (ovp == NULL) {
2591 		/*
2592 		 * When renaming directories to be a subdirectory of a
2593 		 * different parent, the dnlc entry for ".." will no
2594 		 * longer be valid, so it must be removed.
2595 		 *
2596 		 * We do a lookup here to determine whether we are renaming
2597 		 * a directory and we need to check if we are renaming
2598 		 * an unlinked file.  This might have already been done
2599 		 * in previous code, so we check ovp == NULL to avoid
2600 		 * doing it twice.
2601 		 */
2602 
2603 		error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0);
2604 
2605 		/*
2606 		 * The source name *should* already exist.
2607 		 */
2608 		if (error) {
2609 			nfs_rw_exit(&odrp->r_rwlock);
2610 			nfs_rw_exit(&ndrp->r_rwlock);
2611 			if (nvp) {
2612 				VN_RELE(nvp);
2613 			}
2614 			return (error);
2615 		}
2616 		ASSERT(ovp != NULL);
2617 	}
2618 
2619 	dnlc_remove(odvp, onm);
2620 	dnlc_remove(ndvp, nnm);
2621 
2622 	setdiropargs(&args.rna_from, onm, odvp);
2623 	setdiropargs(&args.rna_to, nnm, ndvp);
2624 
2625 	douprintf = 1;
2626 
2627 	error = rfs2call(VTOMI(odvp), RFS_RENAME,
2628 	    xdr_rnmargs, (caddr_t)&args,
2629 	    xdr_enum, (caddr_t)&status, cr,
2630 	    &douprintf, &status, 0, NULL);
2631 
2632 	PURGE_ATTRCACHE(odvp);	/* mod time changed */
2633 	PURGE_ATTRCACHE(ndvp);	/* mod time changed */
2634 
2635 	if (!error) {
2636 		error = geterrno(status);
2637 		if (!error) {
2638 			if (HAVE_RDDIR_CACHE(odrp))
2639 				nfs_purge_rddir_cache(odvp);
2640 			if (HAVE_RDDIR_CACHE(ndrp))
2641 				nfs_purge_rddir_cache(ndvp);
2642 			/*
2643 			 * when renaming directories to be a subdirectory of a
2644 			 * different parent, the dnlc entry for ".." will no
2645 			 * longer be valid, so it must be removed
2646 			 */
2647 			rp = VTOR(ovp);
2648 			if (ndvp != odvp) {
2649 				if (ovp->v_type == VDIR) {
2650 					dnlc_remove(ovp, "..");
2651 					if (HAVE_RDDIR_CACHE(rp))
2652 						nfs_purge_rddir_cache(ovp);
2653 				}
2654 			}
2655 
2656 			/*
2657 			 * If we are renaming the unlinked file, update the
2658 			 * r_unldvp and r_unlname as needed.
2659 			 */
2660 			mutex_enter(&rp->r_statelock);
2661 			if (rp->r_unldvp != NULL) {
2662 				if (strcmp(rp->r_unlname, onm) == 0) {
2663 					(void) strncpy(rp->r_unlname,
2664 					    nnm, MAXNAMELEN);
2665 					rp->r_unlname[MAXNAMELEN - 1] = '\0';
2666 
2667 					if (ndvp != rp->r_unldvp) {
2668 						VN_RELE(rp->r_unldvp);
2669 						rp->r_unldvp = ndvp;
2670 						VN_HOLD(ndvp);
2671 					}
2672 				}
2673 			}
2674 			mutex_exit(&rp->r_statelock);
2675 		} else {
2676 			/*
2677 			 * System V defines rename to return EEXIST, not
2678 			 * ENOTEMPTY if the target directory is not empty.
2679 			 * Over the wire, the error is NFSERR_ENOTEMPTY
2680 			 * which geterrno maps to ENOTEMPTY.
2681 			 */
2682 			if (error == ENOTEMPTY)
2683 				error = EEXIST;
2684 		}
2685 	}
2686 
2687 	if (error == 0) {
2688 		if (nvp)
2689 			vnevent_rename_dest(nvp, ndvp, nnm, ct);
2690 
2691 		if (odvp != ndvp)
2692 			vnevent_rename_dest_dir(ndvp, ct);
2693 
2694 		ASSERT(ovp != NULL);
2695 		vnevent_rename_src(ovp, odvp, onm, ct);
2696 	}
2697 
2698 	if (nvp) {
2699 		VN_RELE(nvp);
2700 	}
2701 	VN_RELE(ovp);
2702 
2703 	nfs_rw_exit(&odrp->r_rwlock);
2704 	nfs_rw_exit(&ndrp->r_rwlock);
2705 
2706 	return (error);
2707 }
2708 
2709 /* ARGSUSED */
2710 static int
2711 nfs_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr,
2712 	caller_context_t *ct, int flags, vsecattr_t *vsecp)
2713 {
2714 	int error;
2715 	struct nfscreatargs args;
2716 	struct nfsdiropres dr;
2717 	int douprintf;
2718 	rnode_t *drp;
2719 	hrtime_t t;
2720 
2721 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2722 		return (EPERM);
2723 
2724 	setdiropargs(&args.ca_da, nm, dvp);
2725 
2726 	/*
2727 	 * Decide what the group-id and set-gid bit of the created directory
2728 	 * should be.  May have to do a setattr to get the gid right.
2729 	 */
2730 	error = setdirgid(dvp, &va->va_gid, cr);
2731 	if (error)
2732 		return (error);
2733 	error = setdirmode(dvp, &va->va_mode, cr);
2734 	if (error)
2735 		return (error);
2736 	va->va_mask |= AT_MODE|AT_GID;
2737 
2738 	args.ca_sa = &args.ca_sa_buf;
2739 	error = vattr_to_sattr(va, args.ca_sa);
2740 	if (error) {
2741 		/* req time field(s) overflow - return immediately */
2742 		return (error);
2743 	}
2744 
2745 	drp = VTOR(dvp);
2746 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2747 		return (EINTR);
2748 
2749 	dnlc_remove(dvp, nm);
2750 
2751 	douprintf = 1;
2752 
2753 	t = gethrtime();
2754 
2755 	error = rfs2call(VTOMI(dvp), RFS_MKDIR,
2756 	    xdr_creatargs, (caddr_t)&args,
2757 	    xdr_diropres, (caddr_t)&dr, cr,
2758 	    &douprintf, &dr.dr_status, 0, NULL);
2759 
2760 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2761 
2762 	if (!error) {
2763 		error = geterrno(dr.dr_status);
2764 		if (!error) {
2765 			if (HAVE_RDDIR_CACHE(drp))
2766 				nfs_purge_rddir_cache(dvp);
2767 			/*
2768 			 * The attributes returned by RFS_MKDIR can not
2769 			 * be depended upon, so mark the attribute cache
2770 			 * as purged.  A subsequent GETATTR will get the
2771 			 * correct attributes from the server.
2772 			 */
2773 			*vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2774 			    dvp->v_vfsp, t, cr, NULL, NULL);
2775 			PURGE_ATTRCACHE(*vpp);
2776 			dnlc_update(dvp, nm, *vpp);
2777 
2778 			/*
2779 			 * Make sure the gid was set correctly.
2780 			 * If not, try to set it (but don't lose
2781 			 * any sleep over it).
2782 			 */
2783 			if (va->va_gid != VTOR(*vpp)->r_attr.va_gid) {
2784 				va->va_mask = AT_GID;
2785 				(void) nfssetattr(*vpp, va, 0, cr);
2786 			}
2787 		} else {
2788 			PURGE_STALE_FH(error, dvp, cr);
2789 		}
2790 	}
2791 
2792 	nfs_rw_exit(&drp->r_rwlock);
2793 
2794 	return (error);
2795 }
2796 
2797 /* ARGSUSED */
2798 static int
2799 nfs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
2800 	caller_context_t *ct, int flags)
2801 {
2802 	int error;
2803 	enum nfsstat status;
2804 	struct nfsdiropargs da;
2805 	vnode_t *vp;
2806 	int douprintf;
2807 	rnode_t *drp;
2808 
2809 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2810 		return (EPERM);
2811 	drp = VTOR(dvp);
2812 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2813 		return (EINTR);
2814 
2815 	/*
2816 	 * Attempt to prevent a rmdir(".") from succeeding.
2817 	 */
2818 	error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2819 	if (error) {
2820 		nfs_rw_exit(&drp->r_rwlock);
2821 		return (error);
2822 	}
2823 
2824 	if (vp == cdir) {
2825 		VN_RELE(vp);
2826 		nfs_rw_exit(&drp->r_rwlock);
2827 		return (EINVAL);
2828 	}
2829 
2830 	setdiropargs(&da, nm, dvp);
2831 
2832 	/*
2833 	 * First just remove the entry from the name cache, as it
2834 	 * is most likely an entry for this vp.
2835 	 */
2836 	dnlc_remove(dvp, nm);
2837 
2838 	/*
2839 	 * If there vnode reference count is greater than one, then
2840 	 * there may be additional references in the DNLC which will
2841 	 * need to be purged.  First, trying removing the entry for
2842 	 * the parent directory and see if that removes the additional
2843 	 * reference(s).  If that doesn't do it, then use dnlc_purge_vp
2844 	 * to completely remove any references to the directory which
2845 	 * might still exist in the DNLC.
2846 	 */
2847 	if (vp->v_count > 1) {
2848 		dnlc_remove(vp, "..");
2849 		if (vp->v_count > 1)
2850 			dnlc_purge_vp(vp);
2851 	}
2852 
2853 	douprintf = 1;
2854 
2855 	error = rfs2call(VTOMI(dvp), RFS_RMDIR,
2856 	    xdr_diropargs, (caddr_t)&da,
2857 	    xdr_enum, (caddr_t)&status, cr,
2858 	    &douprintf, &status, 0, NULL);
2859 
2860 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2861 
2862 	if (error) {
2863 		VN_RELE(vp);
2864 		nfs_rw_exit(&drp->r_rwlock);
2865 		return (error);
2866 	}
2867 
2868 	error = geterrno(status);
2869 	if (!error) {
2870 		if (HAVE_RDDIR_CACHE(drp))
2871 			nfs_purge_rddir_cache(dvp);
2872 		if (HAVE_RDDIR_CACHE(VTOR(vp)))
2873 			nfs_purge_rddir_cache(vp);
2874 	} else {
2875 		PURGE_STALE_FH(error, dvp, cr);
2876 		/*
2877 		 * System V defines rmdir to return EEXIST, not
2878 		 * ENOTEMPTY if the directory is not empty.  Over
2879 		 * the wire, the error is NFSERR_ENOTEMPTY which
2880 		 * geterrno maps to ENOTEMPTY.
2881 		 */
2882 		if (error == ENOTEMPTY)
2883 			error = EEXIST;
2884 	}
2885 
2886 	if (error == 0) {
2887 		vnevent_rmdir(vp, dvp, nm, ct);
2888 	}
2889 	VN_RELE(vp);
2890 
2891 	nfs_rw_exit(&drp->r_rwlock);
2892 
2893 	return (error);
2894 }
2895 
2896 /* ARGSUSED */
2897 static int
2898 nfs_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr,
2899 	caller_context_t *ct, int flags)
2900 {
2901 	int error;
2902 	struct nfsslargs args;
2903 	enum nfsstat status;
2904 	int douprintf;
2905 	rnode_t *drp;
2906 
2907 	if (nfs_zone() != VTOMI(dvp)->mi_zone)
2908 		return (EPERM);
2909 	setdiropargs(&args.sla_from, lnm, dvp);
2910 	args.sla_sa = &args.sla_sa_buf;
2911 	error = vattr_to_sattr(tva, args.sla_sa);
2912 	if (error) {
2913 		/* req time field(s) overflow - return immediately */
2914 		return (error);
2915 	}
2916 	args.sla_tnm = tnm;
2917 
2918 	drp = VTOR(dvp);
2919 	if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2920 		return (EINTR);
2921 
2922 	dnlc_remove(dvp, lnm);
2923 
2924 	douprintf = 1;
2925 
2926 	error = rfs2call(VTOMI(dvp), RFS_SYMLINK,
2927 	    xdr_slargs, (caddr_t)&args,
2928 	    xdr_enum, (caddr_t)&status, cr,
2929 	    &douprintf, &status, 0, NULL);
2930 
2931 	PURGE_ATTRCACHE(dvp);	/* mod time changed */
2932 
2933 	if (!error) {
2934 		error = geterrno(status);
2935 		if (!error) {
2936 			if (HAVE_RDDIR_CACHE(drp))
2937 				nfs_purge_rddir_cache(dvp);
2938 		} else {
2939 			PURGE_STALE_FH(error, dvp, cr);
2940 		}
2941 	}
2942 
2943 	nfs_rw_exit(&drp->r_rwlock);
2944 
2945 	return (error);
2946 }
2947 
2948 #ifdef DEBUG
2949 static int nfs_readdir_cache_hits = 0;
2950 static int nfs_readdir_cache_shorts = 0;
2951 static int nfs_readdir_cache_waits = 0;
2952 static int nfs_readdir_cache_misses = 0;
2953 static int nfs_readdir_readahead = 0;
2954 #endif
2955 
2956 static int nfs_shrinkreaddir = 0;
2957 
2958 /*
2959  * Read directory entries.
2960  * There are some weird things to look out for here.  The uio_offset
2961  * field is either 0 or it is the offset returned from a previous
2962  * readdir.  It is an opaque value used by the server to find the
2963  * correct directory block to read. The count field is the number
2964  * of blocks to read on the server.  This is advisory only, the server
2965  * may return only one block's worth of entries.  Entries may be compressed
2966  * on the server.
2967  */
2968 /* ARGSUSED */
2969 static int
2970 nfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
2971 	caller_context_t *ct, int flags)
2972 {
2973 	int error;
2974 	size_t count;
2975 	rnode_t *rp;
2976 	rddir_cache *rdc;
2977 	rddir_cache *nrdc;
2978 	rddir_cache *rrdc;
2979 #ifdef DEBUG
2980 	int missed;
2981 #endif
2982 	rddir_cache srdc;
2983 	avl_index_t where;
2984 
2985 	rp = VTOR(vp);
2986 
2987 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
2988 	if (nfs_zone() != VTOMI(vp)->mi_zone)
2989 		return (EIO);
2990 	/*
2991 	 * Make sure that the directory cache is valid.
2992 	 */
2993 	if (HAVE_RDDIR_CACHE(rp)) {
2994 		if (nfs_disable_rddir_cache) {
2995 			/*
2996 			 * Setting nfs_disable_rddir_cache in /etc/system
2997 			 * allows interoperability with servers that do not
2998 			 * properly update the attributes of directories.
2999 			 * Any cached information gets purged before an
3000 			 * access is made to it.
3001 			 */
3002 			nfs_purge_rddir_cache(vp);
3003 		} else {
3004 			error = nfs_validate_caches(vp, cr);
3005 			if (error)
3006 				return (error);
3007 		}
3008 	}
3009 
3010 	/*
3011 	 * UGLINESS: SunOS 3.2 servers apparently cannot always handle an
3012 	 * RFS_READDIR request with rda_count set to more than 0x400. So
3013 	 * we reduce the request size here purely for compatibility.
3014 	 *
3015 	 * In general, this is no longer required.  However, if a server
3016 	 * is discovered which can not handle requests larger than 1024,
3017 	 * nfs_shrinkreaddir can be set to 1 to enable this backwards
3018 	 * compatibility.
3019 	 *
3020 	 * In any case, the request size is limited to NFS_MAXDATA bytes.
3021 	 */
3022 	count = MIN(uiop->uio_iov->iov_len,
3023 	    nfs_shrinkreaddir ? 0x400 : NFS_MAXDATA);
3024 
3025 	nrdc = NULL;
3026 #ifdef DEBUG
3027 	missed = 0;
3028 #endif
3029 top:
3030 	/*
3031 	 * Short circuit last readdir which always returns 0 bytes.
3032 	 * This can be done after the directory has been read through
3033 	 * completely at least once.  This will set r_direof which
3034 	 * can be used to find the value of the last cookie.
3035 	 */
3036 	mutex_enter(&rp->r_statelock);
3037 	if (rp->r_direof != NULL &&
3038 	    uiop->uio_offset == rp->r_direof->nfs_ncookie) {
3039 		mutex_exit(&rp->r_statelock);
3040 #ifdef DEBUG
3041 		nfs_readdir_cache_shorts++;
3042 #endif
3043 		if (eofp)
3044 			*eofp = 1;
3045 		if (nrdc != NULL)
3046 			rddir_cache_rele(nrdc);
3047 		return (0);
3048 	}
3049 	/*
3050 	 * Look for a cache entry.  Cache entries are identified
3051 	 * by the NFS cookie value and the byte count requested.
3052 	 */
3053 	srdc.nfs_cookie = uiop->uio_offset;
3054 	srdc.buflen = count;
3055 	rdc = avl_find(&rp->r_dir, &srdc, &where);
3056 	if (rdc != NULL) {
3057 		rddir_cache_hold(rdc);
3058 		/*
3059 		 * If the cache entry is in the process of being
3060 		 * filled in, wait until this completes.  The
3061 		 * RDDIRWAIT bit is set to indicate that someone
3062 		 * is waiting and then the thread currently
3063 		 * filling the entry is done, it should do a
3064 		 * cv_broadcast to wakeup all of the threads
3065 		 * waiting for it to finish.
3066 		 */
3067 		if (rdc->flags & RDDIR) {
3068 			nfs_rw_exit(&rp->r_rwlock);
3069 			rdc->flags |= RDDIRWAIT;
3070 #ifdef DEBUG
3071 			nfs_readdir_cache_waits++;
3072 #endif
3073 			if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) {
3074 				/*
3075 				 * We got interrupted, probably
3076 				 * the user typed ^C or an alarm
3077 				 * fired.  We free the new entry
3078 				 * if we allocated one.
3079 				 */
3080 				mutex_exit(&rp->r_statelock);
3081 				(void) nfs_rw_enter_sig(&rp->r_rwlock,
3082 				    RW_READER, FALSE);
3083 				rddir_cache_rele(rdc);
3084 				if (nrdc != NULL)
3085 					rddir_cache_rele(nrdc);
3086 				return (EINTR);
3087 			}
3088 			mutex_exit(&rp->r_statelock);
3089 			(void) nfs_rw_enter_sig(&rp->r_rwlock,
3090 			    RW_READER, FALSE);
3091 			rddir_cache_rele(rdc);
3092 			goto top;
3093 		}
3094 		/*
3095 		 * Check to see if a readdir is required to
3096 		 * fill the entry.  If so, mark this entry
3097 		 * as being filled, remove our reference,
3098 		 * and branch to the code to fill the entry.
3099 		 */
3100 		if (rdc->flags & RDDIRREQ) {
3101 			rdc->flags &= ~RDDIRREQ;
3102 			rdc->flags |= RDDIR;
3103 			if (nrdc != NULL)
3104 				rddir_cache_rele(nrdc);
3105 			nrdc = rdc;
3106 			mutex_exit(&rp->r_statelock);
3107 			goto bottom;
3108 		}
3109 #ifdef DEBUG
3110 		if (!missed)
3111 			nfs_readdir_cache_hits++;
3112 #endif
3113 		/*
3114 		 * If an error occurred while attempting
3115 		 * to fill the cache entry, just return it.
3116 		 */
3117 		if (rdc->error) {
3118 			error = rdc->error;
3119 			mutex_exit(&rp->r_statelock);
3120 			rddir_cache_rele(rdc);
3121 			if (nrdc != NULL)
3122 				rddir_cache_rele(nrdc);
3123 			return (error);
3124 		}
3125 
3126 		/*
3127 		 * The cache entry is complete and good,
3128 		 * copyout the dirent structs to the calling
3129 		 * thread.
3130 		 */
3131 		error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop);
3132 
3133 		/*
3134 		 * If no error occurred during the copyout,
3135 		 * update the offset in the uio struct to
3136 		 * contain the value of the next cookie
3137 		 * and set the eof value appropriately.
3138 		 */
3139 		if (!error) {
3140 			uiop->uio_offset = rdc->nfs_ncookie;
3141 			if (eofp)
3142 				*eofp = rdc->eof;
3143 		}
3144 
3145 		/*
3146 		 * Decide whether to do readahead.  Don't if
3147 		 * have already read to the end of directory.
3148 		 */
3149 		if (rdc->eof) {
3150 			rp->r_direof = rdc;
3151 			mutex_exit(&rp->r_statelock);
3152 			rddir_cache_rele(rdc);
3153 			if (nrdc != NULL)
3154 				rddir_cache_rele(nrdc);
3155 			return (error);
3156 		}
3157 
3158 		/*
3159 		 * Check to see whether we found an entry
3160 		 * for the readahead.  If so, we don't need
3161 		 * to do anything further, so free the new
3162 		 * entry if one was allocated.  Otherwise,
3163 		 * allocate a new entry, add it to the cache,
3164 		 * and then initiate an asynchronous readdir
3165 		 * operation to fill it.
3166 		 */
3167 		srdc.nfs_cookie = rdc->nfs_ncookie;
3168 		srdc.buflen = count;
3169 		rrdc = avl_find(&rp->r_dir, &srdc, &where);
3170 		if (rrdc != NULL) {
3171 			if (nrdc != NULL)
3172 				rddir_cache_rele(nrdc);
3173 		} else {
3174 			if (nrdc != NULL)
3175 				rrdc = nrdc;
3176 			else {
3177 				rrdc = rddir_cache_alloc(KM_NOSLEEP);
3178 			}
3179 			if (rrdc != NULL) {
3180 				rrdc->nfs_cookie = rdc->nfs_ncookie;
3181 				rrdc->buflen = count;
3182 				avl_insert(&rp->r_dir, rrdc, where);
3183 				rddir_cache_hold(rrdc);
3184 				mutex_exit(&rp->r_statelock);
3185 				rddir_cache_rele(rdc);
3186 #ifdef DEBUG
3187 				nfs_readdir_readahead++;
3188 #endif
3189 				nfs_async_readdir(vp, rrdc, cr, nfsreaddir);
3190 				return (error);
3191 			}
3192 		}
3193 
3194 		mutex_exit(&rp->r_statelock);
3195 		rddir_cache_rele(rdc);
3196 		return (error);
3197 	}
3198 
3199 	/*
3200 	 * Didn't find an entry in the cache.  Construct a new empty
3201 	 * entry and link it into the cache.  Other processes attempting
3202 	 * to access this entry will need to wait until it is filled in.
3203 	 *
3204 	 * Since kmem_alloc may block, another pass through the cache
3205 	 * will need to be taken to make sure that another process
3206 	 * hasn't already added an entry to the cache for this request.
3207 	 */
3208 	if (nrdc == NULL) {
3209 		mutex_exit(&rp->r_statelock);
3210 		nrdc = rddir_cache_alloc(KM_SLEEP);
3211 		nrdc->nfs_cookie = uiop->uio_offset;
3212 		nrdc->buflen = count;
3213 		goto top;
3214 	}
3215 
3216 	/*
3217 	 * Add this entry to the cache.
3218 	 */
3219 	avl_insert(&rp->r_dir, nrdc, where);
3220 	rddir_cache_hold(nrdc);
3221 	mutex_exit(&rp->r_statelock);
3222 
3223 bottom:
3224 #ifdef DEBUG
3225 	missed = 1;
3226 	nfs_readdir_cache_misses++;
3227 #endif
3228 	/*
3229 	 * Do the readdir.
3230 	 */
3231 	error = nfsreaddir(vp, nrdc, cr);
3232 
3233 	/*
3234 	 * If this operation failed, just return the error which occurred.
3235 	 */
3236 	if (error != 0)
3237 		return (error);
3238 
3239 	/*
3240 	 * Since the RPC operation will have taken sometime and blocked
3241 	 * this process, another pass through the cache will need to be
3242 	 * taken to find the correct cache entry.  It is possible that
3243 	 * the correct cache entry will not be there (although one was
3244 	 * added) because the directory changed during the RPC operation
3245 	 * and the readdir cache was flushed.  In this case, just start
3246 	 * over.  It is hoped that this will not happen too often... :-)
3247 	 */
3248 	nrdc = NULL;
3249 	goto top;
3250 	/* NOTREACHED */
3251 }
3252 
3253 static int
3254 nfsreaddir(vnode_t *vp, rddir_cache *rdc, cred_t *cr)
3255 {
3256 	int error;
3257 	struct nfsrddirargs rda;
3258 	struct nfsrddirres rd;
3259 	rnode_t *rp;
3260 	mntinfo_t *mi;
3261 	uint_t count;
3262 	int douprintf;
3263 	failinfo_t fi, *fip;
3264 
3265 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3266 	count = rdc->buflen;
3267 
3268 	rp = VTOR(vp);
3269 	mi = VTOMI(vp);
3270 
3271 	rda.rda_fh = *VTOFH(vp);
3272 	rda.rda_offset = rdc->nfs_cookie;
3273 
3274 	/*
3275 	 * NFS client failover support
3276 	 * suppress failover unless we have a zero cookie
3277 	 */
3278 	if (rdc->nfs_cookie == (off_t)0) {
3279 		fi.vp = vp;
3280 		fi.fhp = (caddr_t)&rda.rda_fh;
3281 		fi.copyproc = nfscopyfh;
3282 		fi.lookupproc = nfslookup;
3283 		fi.xattrdirproc = acl_getxattrdir2;
3284 		fip = &fi;
3285 	} else {
3286 		fip = NULL;
3287 	}
3288 
3289 	rd.rd_entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3290 	rd.rd_size = count;
3291 	rd.rd_offset = rda.rda_offset;
3292 
3293 	douprintf = 1;
3294 
3295 	if (mi->mi_io_kstats) {
3296 		mutex_enter(&mi->mi_lock);
3297 		kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
3298 		mutex_exit(&mi->mi_lock);
3299 	}
3300 
3301 	do {
3302 		rda.rda_count = MIN(count, mi->mi_curread);
3303 		error = rfs2call(mi, RFS_READDIR,
3304 		    xdr_rddirargs, (caddr_t)&rda,
3305 		    xdr_getrddirres, (caddr_t)&rd, cr,
3306 		    &douprintf, &rd.rd_status, 0, fip);
3307 	} while (error == ENFS_TRYAGAIN);
3308 
3309 	if (mi->mi_io_kstats) {
3310 		mutex_enter(&mi->mi_lock);
3311 		kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
3312 		mutex_exit(&mi->mi_lock);
3313 	}
3314 
3315 	/*
3316 	 * Since we are actually doing a READDIR RPC, we must have
3317 	 * exclusive access to the cache entry being filled.  Thus,
3318 	 * it is safe to update all fields except for the flags
3319 	 * field.  The r_statelock in the rnode must be held to
3320 	 * prevent two different threads from simultaneously
3321 	 * attempting to update the flags field.  This can happen
3322 	 * if we are turning off RDDIR and the other thread is
3323 	 * trying to set RDDIRWAIT.
3324 	 */
3325 	ASSERT(rdc->flags & RDDIR);
3326 	if (!error) {
3327 		error = geterrno(rd.rd_status);
3328 		if (!error) {
3329 			rdc->nfs_ncookie = rd.rd_offset;
3330 			rdc->eof = rd.rd_eof ? 1 : 0;
3331 			rdc->entlen = rd.rd_size;
3332 			ASSERT(rdc->entlen <= rdc->buflen);
3333 #ifdef DEBUG
3334 			rdc->entries = rddir_cache_buf_alloc(rdc->buflen,
3335 			    KM_SLEEP);
3336 #else
3337 			rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3338 #endif
3339 			bcopy(rd.rd_entries, rdc->entries, rdc->entlen);
3340 			rdc->error = 0;
3341 			if (mi->mi_io_kstats) {
3342 				mutex_enter(&mi->mi_lock);
3343 				KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
3344 				KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
3345 				    rd.rd_size;
3346 				mutex_exit(&mi->mi_lock);
3347 			}
3348 		} else {
3349 			PURGE_STALE_FH(error, vp, cr);
3350 		}
3351 	}
3352 	if (error) {
3353 		rdc->entries = NULL;
3354 		rdc->error = error;
3355 	}
3356 	kmem_free(rd.rd_entries, rdc->buflen);
3357 
3358 	mutex_enter(&rp->r_statelock);
3359 	rdc->flags &= ~RDDIR;
3360 	if (rdc->flags & RDDIRWAIT) {
3361 		rdc->flags &= ~RDDIRWAIT;
3362 		cv_broadcast(&rdc->cv);
3363 	}
3364 	if (error)
3365 		rdc->flags |= RDDIRREQ;
3366 	mutex_exit(&rp->r_statelock);
3367 
3368 	rddir_cache_rele(rdc);
3369 
3370 	return (error);
3371 }
3372 
3373 #ifdef DEBUG
3374 static int nfs_bio_do_stop = 0;
3375 #endif
3376 
3377 static int
3378 nfs_bio(struct buf *bp, cred_t *cr)
3379 {
3380 	rnode_t *rp = VTOR(bp->b_vp);
3381 	int count;
3382 	int error;
3383 	cred_t *cred;
3384 	uint_t offset;
3385 
3386 	DTRACE_IO1(start, struct buf *, bp);
3387 
3388 	ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone);
3389 	offset = dbtob(bp->b_blkno);
3390 
3391 	if (bp->b_flags & B_READ) {
3392 		mutex_enter(&rp->r_statelock);
3393 		if (rp->r_cred != NULL) {
3394 			cred = rp->r_cred;
3395 			crhold(cred);
3396 		} else {
3397 			rp->r_cred = cr;
3398 			crhold(cr);
3399 			cred = cr;
3400 			crhold(cred);
3401 		}
3402 		mutex_exit(&rp->r_statelock);
3403 	read_again:
3404 		error = bp->b_error = nfsread(bp->b_vp, bp->b_un.b_addr,
3405 		    offset, bp->b_bcount, &bp->b_resid, cred);
3406 
3407 		crfree(cred);
3408 		if (!error) {
3409 			if (bp->b_resid) {
3410 				/*
3411 				 * Didn't get it all because we hit EOF,
3412 				 * zero all the memory beyond the EOF.
3413 				 */
3414 				/* bzero(rdaddr + */
3415 				bzero(bp->b_un.b_addr +
3416 				    bp->b_bcount - bp->b_resid, bp->b_resid);
3417 			}
3418 			mutex_enter(&rp->r_statelock);
3419 			if (bp->b_resid == bp->b_bcount &&
3420 			    offset >= rp->r_size) {
3421 				/*
3422 				 * We didn't read anything at all as we are
3423 				 * past EOF.  Return an error indicator back
3424 				 * but don't destroy the pages (yet).
3425 				 */
3426 				error = NFS_EOF;
3427 			}
3428 			mutex_exit(&rp->r_statelock);
3429 		} else if (error == EACCES) {
3430 			mutex_enter(&rp->r_statelock);
3431 			if (cred != cr) {
3432 				if (rp->r_cred != NULL)
3433 					crfree(rp->r_cred);
3434 				rp->r_cred = cr;
3435 				crhold(cr);
3436 				cred = cr;
3437 				crhold(cred);
3438 				mutex_exit(&rp->r_statelock);
3439 				goto read_again;
3440 			}
3441 			mutex_exit(&rp->r_statelock);
3442 		}
3443 	} else {
3444 		if (!(rp->r_flags & RSTALE)) {
3445 			mutex_enter(&rp->r_statelock);
3446 			if (rp->r_cred != NULL) {
3447 				cred = rp->r_cred;
3448 				crhold(cred);
3449 			} else {
3450 				rp->r_cred = cr;
3451 				crhold(cr);
3452 				cred = cr;
3453 				crhold(cred);
3454 			}
3455 			mutex_exit(&rp->r_statelock);
3456 		write_again:
3457 			mutex_enter(&rp->r_statelock);
3458 			count = MIN(bp->b_bcount, rp->r_size - offset);
3459 			mutex_exit(&rp->r_statelock);
3460 			if (count < 0)
3461 				cmn_err(CE_PANIC, "nfs_bio: write count < 0");
3462 #ifdef DEBUG
3463 			if (count == 0) {
3464 				zcmn_err(getzoneid(), CE_WARN,
3465 				    "nfs_bio: zero length write at %d",
3466 				    offset);
3467 				nfs_printfhandle(&rp->r_fh);
3468 				if (nfs_bio_do_stop)
3469 					debug_enter("nfs_bio");
3470 			}
3471 #endif
3472 			error = nfswrite(bp->b_vp, bp->b_un.b_addr, offset,
3473 			    count, cred);
3474 			if (error == EACCES) {
3475 				mutex_enter(&rp->r_statelock);
3476 				if (cred != cr) {
3477 					if (rp->r_cred != NULL)
3478 						crfree(rp->r_cred);
3479 					rp->r_cred = cr;
3480 					crhold(cr);
3481 					crfree(cred);
3482 					cred = cr;
3483 					crhold(cred);
3484 					mutex_exit(&rp->r_statelock);
3485 					goto write_again;
3486 				}
3487 				mutex_exit(&rp->r_statelock);
3488 			}
3489 			bp->b_error = error;
3490 			if (error && error != EINTR) {
3491 				/*
3492 				 * Don't print EDQUOT errors on the console.
3493 				 * Don't print asynchronous EACCES errors.
3494 				 * Don't print EFBIG errors.
3495 				 * Print all other write errors.
3496 				 */
3497 				if (error != EDQUOT && error != EFBIG &&
3498 				    (error != EACCES ||
3499 				    !(bp->b_flags & B_ASYNC)))
3500 					nfs_write_error(bp->b_vp, error, cred);
3501 				/*
3502 				 * Update r_error and r_flags as appropriate.
3503 				 * If the error was ESTALE, then mark the
3504 				 * rnode as not being writeable and save
3505 				 * the error status.  Otherwise, save any
3506 				 * errors which occur from asynchronous
3507 				 * page invalidations.  Any errors occurring
3508 				 * from other operations should be saved
3509 				 * by the caller.
3510 				 */
3511 				mutex_enter(&rp->r_statelock);
3512 				if (error == ESTALE) {
3513 					rp->r_flags |= RSTALE;
3514 					if (!rp->r_error)
3515 						rp->r_error = error;
3516 				} else if (!rp->r_error &&
3517 				    (bp->b_flags &
3518 				    (B_INVAL|B_FORCE|B_ASYNC)) ==
3519 				    (B_INVAL|B_FORCE|B_ASYNC)) {
3520 					rp->r_error = error;
3521 				}
3522 				mutex_exit(&rp->r_statelock);
3523 			}
3524 			crfree(cred);
3525 		} else {
3526 			error = rp->r_error;
3527 			/*
3528 			 * A close may have cleared r_error, if so,
3529 			 * propagate ESTALE error return properly
3530 			 */
3531 			if (error == 0)
3532 				error = ESTALE;
3533 		}
3534 	}
3535 
3536 	if (error != 0 && error != NFS_EOF)
3537 		bp->b_flags |= B_ERROR;
3538 
3539 	DTRACE_IO1(done, struct buf *, bp);
3540 
3541 	return (error);
3542 }
3543 
3544 /* ARGSUSED */
3545 static int
3546 nfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3547 {
3548 	struct nfs_fid *fp;
3549 	rnode_t *rp;
3550 
3551 	rp = VTOR(vp);
3552 
3553 	if (fidp->fid_len < (sizeof (struct nfs_fid) - sizeof (short))) {
3554 		fidp->fid_len = sizeof (struct nfs_fid) - sizeof (short);
3555 		return (ENOSPC);
3556 	}
3557 	fp = (struct nfs_fid *)fidp;
3558 	fp->nf_pad = 0;
3559 	fp->nf_len = sizeof (struct nfs_fid) - sizeof (short);
3560 	bcopy(rp->r_fh.fh_buf, fp->nf_data, NFS_FHSIZE);
3561 	return (0);
3562 }
3563 
3564 /* ARGSUSED2 */
3565 static int
3566 nfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3567 {
3568 	rnode_t *rp = VTOR(vp);
3569 
3570 	if (!write_lock) {
3571 		(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3572 		return (V_WRITELOCK_FALSE);
3573 	}
3574 
3575 	if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) {
3576 		(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3577 		if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp))
3578 			return (V_WRITELOCK_FALSE);
3579 		nfs_rw_exit(&rp->r_rwlock);
3580 	}
3581 
3582 	(void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE);
3583 	return (V_WRITELOCK_TRUE);
3584 }
3585 
3586 /* ARGSUSED */
3587 static void
3588 nfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3589 {
3590 	rnode_t *rp = VTOR(vp);
3591 
3592 	nfs_rw_exit(&rp->r_rwlock);
3593 }
3594 
3595 /* ARGSUSED */
3596 static int
3597 nfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
3598 {
3599 
3600 	/*
3601 	 * Because we stuff the readdir cookie into the offset field
3602 	 * someone may attempt to do an lseek with the cookie which
3603 	 * we want to succeed.
3604 	 */
3605 	if (vp->v_type == VDIR)
3606 		return (0);
3607 	if (*noffp < 0 || *noffp > MAXOFF32_T)
3608 		return (EINVAL);
3609 	return (0);
3610 }
3611 
3612 /*
3613  * number of NFS_MAXDATA blocks to read ahead
3614  * optimized for 100 base-T.
3615  */
3616 static int nfs_nra = 4;
3617 
3618 #ifdef DEBUG
3619 static int nfs_lostpage = 0;	/* number of times we lost original page */
3620 #endif
3621 
3622 /*
3623  * Return all the pages from [off..off+len) in file
3624  */
3625 /* ARGSUSED */
3626 static int
3627 nfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3628 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3629 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3630 {
3631 	rnode_t *rp;
3632 	int error;
3633 	mntinfo_t *mi;
3634 
3635 	if (vp->v_flag & VNOMAP)
3636 		return (ENOSYS);
3637 
3638 	ASSERT(off <= MAXOFF32_T);
3639 	if (nfs_zone() != VTOMI(vp)->mi_zone)
3640 		return (EIO);
3641 	if (protp != NULL)
3642 		*protp = PROT_ALL;
3643 
3644 	/*
3645 	 * Now valididate that the caches are up to date.
3646 	 */
3647 	error = nfs_validate_caches(vp, cr);
3648 	if (error)
3649 		return (error);
3650 
3651 	rp = VTOR(vp);
3652 	mi = VTOMI(vp);
3653 retry:
3654 	mutex_enter(&rp->r_statelock);
3655 
3656 	/*
3657 	 * Don't create dirty pages faster than they
3658 	 * can be cleaned so that the system doesn't
3659 	 * get imbalanced.  If the async queue is
3660 	 * maxed out, then wait for it to drain before
3661 	 * creating more dirty pages.  Also, wait for
3662 	 * any threads doing pagewalks in the vop_getattr
3663 	 * entry points so that they don't block for
3664 	 * long periods.
3665 	 */
3666 	if (rw == S_CREATE) {
3667 		while ((mi->mi_max_threads != 0 &&
3668 		    rp->r_awcount > 2 * mi->mi_max_threads) ||
3669 		    rp->r_gcount > 0)
3670 			cv_wait(&rp->r_cv, &rp->r_statelock);
3671 	}
3672 
3673 	/*
3674 	 * If we are getting called as a side effect of an nfs_write()
3675 	 * operation the local file size might not be extended yet.
3676 	 * In this case we want to be able to return pages of zeroes.
3677 	 */
3678 	if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) {
3679 		mutex_exit(&rp->r_statelock);
3680 		return (EFAULT);		/* beyond EOF */
3681 	}
3682 
3683 	mutex_exit(&rp->r_statelock);
3684 
3685 	error = pvn_getpages(nfs_getapage, vp, off, len, protp, pl, plsz,
3686 	    seg, addr, rw, cr);
3687 
3688 	switch (error) {
3689 	case NFS_EOF:
3690 		nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr);
3691 		goto retry;
3692 	case ESTALE:
3693 		PURGE_STALE_FH(error, vp, cr);
3694 	}
3695 
3696 	return (error);
3697 }
3698 
3699 /*
3700  * Called from pvn_getpages to get a particular page.
3701  */
3702 /* ARGSUSED */
3703 static int
3704 nfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
3705 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3706 	enum seg_rw rw, cred_t *cr)
3707 {
3708 	rnode_t *rp;
3709 	uint_t bsize;
3710 	struct buf *bp;
3711 	page_t *pp;
3712 	u_offset_t lbn;
3713 	u_offset_t io_off;
3714 	u_offset_t blkoff;
3715 	u_offset_t rablkoff;
3716 	size_t io_len;
3717 	uint_t blksize;
3718 	int error;
3719 	int readahead;
3720 	int readahead_issued = 0;
3721 	int ra_window; /* readahead window */
3722 	page_t *pagefound;
3723 
3724 	if (nfs_zone() != VTOMI(vp)->mi_zone)
3725 		return (EIO);
3726 	rp = VTOR(vp);
3727 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3728 
3729 reread:
3730 	bp = NULL;
3731 	pp = NULL;
3732 	pagefound = NULL;
3733 
3734 	if (pl != NULL)
3735 		pl[0] = NULL;
3736 
3737 	error = 0;
3738 	lbn = off / bsize;
3739 	blkoff = lbn * bsize;
3740 
3741 	/*
3742 	 * Queueing up the readahead before doing the synchronous read
3743 	 * results in a significant increase in read throughput because
3744 	 * of the increased parallelism between the async threads and
3745 	 * the process context.
3746 	 */
3747 	if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 &&
3748 	    rw != S_CREATE &&
3749 	    !(vp->v_flag & VNOCACHE)) {
3750 		mutex_enter(&rp->r_statelock);
3751 
3752 		/*
3753 		 * Calculate the number of readaheads to do.
3754 		 * a) No readaheads at offset = 0.
3755 		 * b) Do maximum(nfs_nra) readaheads when the readahead
3756 		 *    window is closed.
3757 		 * c) Do readaheads between 1 to (nfs_nra - 1) depending
3758 		 *    upon how far the readahead window is open or close.
3759 		 * d) No readaheads if rp->r_nextr is not within the scope
3760 		 *    of the readahead window (random i/o).
3761 		 */
3762 
3763 		if (off == 0)
3764 			readahead = 0;
3765 		else if (blkoff == rp->r_nextr)
3766 			readahead = nfs_nra;
3767 		else if (rp->r_nextr > blkoff &&
3768 		    ((ra_window = (rp->r_nextr - blkoff) / bsize)
3769 		    <= (nfs_nra - 1)))
3770 			readahead = nfs_nra - ra_window;
3771 		else
3772 			readahead = 0;
3773 
3774 		rablkoff = rp->r_nextr;
3775 		while (readahead > 0 && rablkoff + bsize < rp->r_size) {
3776 			mutex_exit(&rp->r_statelock);
3777 			if (nfs_async_readahead(vp, rablkoff + bsize,
3778 			    addr + (rablkoff + bsize - off), seg, cr,
3779 			    nfs_readahead) < 0) {
3780 				mutex_enter(&rp->r_statelock);
3781 				break;
3782 			}
3783 			readahead--;
3784 			rablkoff += bsize;
3785 			/*
3786 			 * Indicate that we did a readahead so
3787 			 * readahead offset is not updated
3788 			 * by the synchronous read below.
3789 			 */
3790 			readahead_issued = 1;
3791 			mutex_enter(&rp->r_statelock);
3792 			/*
3793 			 * set readahead offset to
3794 			 * offset of last async readahead
3795 			 * request.
3796 			 */
3797 			rp->r_nextr = rablkoff;
3798 		}
3799 		mutex_exit(&rp->r_statelock);
3800 	}
3801 
3802 again:
3803 	if ((pagefound = page_exists(vp, off)) == NULL) {
3804 		if (pl == NULL) {
3805 			(void) nfs_async_readahead(vp, blkoff, addr, seg, cr,
3806 			    nfs_readahead);
3807 		} else if (rw == S_CREATE) {
3808 			/*
3809 			 * Block for this page is not allocated, or the offset
3810 			 * is beyond the current allocation size, or we're
3811 			 * allocating a swap slot and the page was not found,
3812 			 * so allocate it and return a zero page.
3813 			 */
3814 			if ((pp = page_create_va(vp, off,
3815 			    PAGESIZE, PG_WAIT, seg, addr)) == NULL)
3816 				cmn_err(CE_PANIC, "nfs_getapage: page_create");
3817 			io_len = PAGESIZE;
3818 			mutex_enter(&rp->r_statelock);
3819 			rp->r_nextr = off + PAGESIZE;
3820 			mutex_exit(&rp->r_statelock);
3821 		} else {
3822 			/*
3823 			 * Need to go to server to get a BLOCK, exception to
3824 			 * that being while reading at offset = 0 or doing
3825 			 * random i/o, in that case read only a PAGE.
3826 			 */
3827 			mutex_enter(&rp->r_statelock);
3828 			if (blkoff < rp->r_size &&
3829 			    blkoff + bsize >= rp->r_size) {
3830 				/*
3831 				 * If only a block or less is left in
3832 				 * the file, read all that is remaining.
3833 				 */
3834 				if (rp->r_size <= off) {
3835 					/*
3836 					 * Trying to access beyond EOF,
3837 					 * set up to get at least one page.
3838 					 */
3839 					blksize = off + PAGESIZE - blkoff;
3840 				} else
3841 					blksize = rp->r_size - blkoff;
3842 			} else if ((off == 0) ||
3843 			    (off != rp->r_nextr && !readahead_issued)) {
3844 				blksize = PAGESIZE;
3845 				blkoff = off; /* block = page here */
3846 			} else
3847 				blksize = bsize;
3848 			mutex_exit(&rp->r_statelock);
3849 
3850 			pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3851 			    &io_len, blkoff, blksize, 0);
3852 
3853 			/*
3854 			 * Some other thread has entered the page,
3855 			 * so just use it.
3856 			 */
3857 			if (pp == NULL)
3858 				goto again;
3859 
3860 			/*
3861 			 * Now round the request size up to page boundaries.
3862 			 * This ensures that the entire page will be
3863 			 * initialized to zeroes if EOF is encountered.
3864 			 */
3865 			io_len = ptob(btopr(io_len));
3866 
3867 			bp = pageio_setup(pp, io_len, vp, B_READ);
3868 			ASSERT(bp != NULL);
3869 
3870 			/*
3871 			 * pageio_setup should have set b_addr to 0.  This
3872 			 * is correct since we want to do I/O on a page
3873 			 * boundary.  bp_mapin will use this addr to calculate
3874 			 * an offset, and then set b_addr to the kernel virtual
3875 			 * address it allocated for us.
3876 			 */
3877 			ASSERT(bp->b_un.b_addr == 0);
3878 
3879 			bp->b_edev = 0;
3880 			bp->b_dev = 0;
3881 			bp->b_lblkno = lbtodb(io_off);
3882 			bp->b_file = vp;
3883 			bp->b_offset = (offset_t)off;
3884 			bp_mapin(bp);
3885 
3886 			/*
3887 			 * If doing a write beyond what we believe is EOF,
3888 			 * don't bother trying to read the pages from the
3889 			 * server, we'll just zero the pages here.  We
3890 			 * don't check that the rw flag is S_WRITE here
3891 			 * because some implementations may attempt a
3892 			 * read access to the buffer before copying data.
3893 			 */
3894 			mutex_enter(&rp->r_statelock);
3895 			if (io_off >= rp->r_size && seg == segkmap) {
3896 				mutex_exit(&rp->r_statelock);
3897 				bzero(bp->b_un.b_addr, io_len);
3898 			} else {
3899 				mutex_exit(&rp->r_statelock);
3900 				error = nfs_bio(bp, cr);
3901 			}
3902 
3903 			/*
3904 			 * Unmap the buffer before freeing it.
3905 			 */
3906 			bp_mapout(bp);
3907 			pageio_done(bp);
3908 
3909 			if (error == NFS_EOF) {
3910 				/*
3911 				 * If doing a write system call just return
3912 				 * zeroed pages, else user tried to get pages
3913 				 * beyond EOF, return error.  We don't check
3914 				 * that the rw flag is S_WRITE here because
3915 				 * some implementations may attempt a read
3916 				 * access to the buffer before copying data.
3917 				 */
3918 				if (seg == segkmap)
3919 					error = 0;
3920 				else
3921 					error = EFAULT;
3922 			}
3923 
3924 			if (!readahead_issued && !error) {
3925 				mutex_enter(&rp->r_statelock);
3926 				rp->r_nextr = io_off + io_len;
3927 				mutex_exit(&rp->r_statelock);
3928 			}
3929 		}
3930 	}
3931 
3932 out:
3933 	if (pl == NULL)
3934 		return (error);
3935 
3936 	if (error) {
3937 		if (pp != NULL)
3938 			pvn_read_done(pp, B_ERROR);
3939 		return (error);
3940 	}
3941 
3942 	if (pagefound) {
3943 		se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED);
3944 
3945 		/*
3946 		 * Page exists in the cache, acquire the appropriate lock.
3947 		 * If this fails, start all over again.
3948 		 */
3949 		if ((pp = page_lookup(vp, off, se)) == NULL) {
3950 #ifdef DEBUG
3951 			nfs_lostpage++;
3952 #endif
3953 			goto reread;
3954 		}
3955 		pl[0] = pp;
3956 		pl[1] = NULL;
3957 		return (0);
3958 	}
3959 
3960 	if (pp != NULL)
3961 		pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3962 
3963 	return (error);
3964 }
3965 
3966 static void
3967 nfs_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg,
3968 	cred_t *cr)
3969 {
3970 	int error;
3971 	page_t *pp;
3972 	u_offset_t io_off;
3973 	size_t io_len;
3974 	struct buf *bp;
3975 	uint_t bsize, blksize;
3976 	rnode_t *rp = VTOR(vp);
3977 
3978 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3979 
3980 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3981 
3982 	mutex_enter(&rp->r_statelock);
3983 	if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) {
3984 		/*
3985 		 * If less than a block left in file read less
3986 		 * than a block.
3987 		 */
3988 		blksize = rp->r_size - blkoff;
3989 	} else
3990 		blksize = bsize;
3991 	mutex_exit(&rp->r_statelock);
3992 
3993 	pp = pvn_read_kluster(vp, blkoff, segkmap, addr,
3994 	    &io_off, &io_len, blkoff, blksize, 1);
3995 	/*
3996 	 * The isra flag passed to the kluster function is 1, we may have
3997 	 * gotten a return value of NULL for a variety of reasons (# of free
3998 	 * pages < minfree, someone entered the page on the vnode etc). In all
3999 	 * cases, we want to punt on the readahead.
4000 	 */
4001 	if (pp == NULL)
4002 		return;
4003 
4004 	/*
4005 	 * Now round the request size up to page boundaries.
4006 	 * This ensures that the entire page will be
4007 	 * initialized to zeroes if EOF is encountered.
4008 	 */
4009 	io_len = ptob(btopr(io_len));
4010 
4011 	bp = pageio_setup(pp, io_len, vp, B_READ);
4012 	ASSERT(bp != NULL);
4013 
4014 	/*
4015 	 * pageio_setup should have set b_addr to 0.  This is correct since
4016 	 * we want to do I/O on a page boundary. bp_mapin() will use this addr
4017 	 * to calculate an offset, and then set b_addr to the kernel virtual
4018 	 * address it allocated for us.
4019 	 */
4020 	ASSERT(bp->b_un.b_addr == 0);
4021 
4022 	bp->b_edev = 0;
4023 	bp->b_dev = 0;
4024 	bp->b_lblkno = lbtodb(io_off);
4025 	bp->b_file = vp;
4026 	bp->b_offset = (offset_t)blkoff;
4027 	bp_mapin(bp);
4028 
4029 	/*
4030 	 * If doing a write beyond what we believe is EOF, don't bother trying
4031 	 * to read the pages from the server, we'll just zero the pages here.
4032 	 * We don't check that the rw flag is S_WRITE here because some
4033 	 * implementations may attempt a read access to the buffer before
4034 	 * copying data.
4035 	 */
4036 	mutex_enter(&rp->r_statelock);
4037 	if (io_off >= rp->r_size && seg == segkmap) {
4038 		mutex_exit(&rp->r_statelock);
4039 		bzero(bp->b_un.b_addr, io_len);
4040 		error = 0;
4041 	} else {
4042 		mutex_exit(&rp->r_statelock);
4043 		error = nfs_bio(bp, cr);
4044 		if (error == NFS_EOF)
4045 			error = 0;
4046 	}
4047 
4048 	/*
4049 	 * Unmap the buffer before freeing it.
4050 	 */
4051 	bp_mapout(bp);
4052 	pageio_done(bp);
4053 
4054 	pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ);
4055 
4056 	/*
4057 	 * In case of error set readahead offset
4058 	 * to the lowest offset.
4059 	 * pvn_read_done() calls VN_DISPOSE to destroy the pages
4060 	 */
4061 	if (error && rp->r_nextr > io_off) {
4062 		mutex_enter(&rp->r_statelock);
4063 		if (rp->r_nextr > io_off)
4064 			rp->r_nextr = io_off;
4065 		mutex_exit(&rp->r_statelock);
4066 	}
4067 }
4068 
4069 /*
4070  * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
4071  * If len == 0, do from off to EOF.
4072  *
4073  * The normal cases should be len == 0 && off == 0 (entire vp list),
4074  * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
4075  * (from pageout).
4076  */
4077 /* ARGSUSED */
4078 static int
4079 nfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4080 	caller_context_t *ct)
4081 {
4082 	int error;
4083 	rnode_t *rp;
4084 
4085 	ASSERT(cr != NULL);
4086 
4087 	/*
4088 	 * XXX - Why should this check be made here?
4089 	 */
4090 	if (vp->v_flag & VNOMAP)
4091 		return (ENOSYS);
4092 
4093 	if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp))
4094 		return (0);
4095 
4096 	if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone)
4097 		return (EIO);
4098 	ASSERT(off <= MAXOFF32_T);
4099 
4100 	rp = VTOR(vp);
4101 	mutex_enter(&rp->r_statelock);
4102 	rp->r_count++;
4103 	mutex_exit(&rp->r_statelock);
4104 	error = nfs_putpages(vp, off, len, flags, cr);
4105 	mutex_enter(&rp->r_statelock);
4106 	rp->r_count--;
4107 	cv_broadcast(&rp->r_cv);
4108 	mutex_exit(&rp->r_statelock);
4109 
4110 	return (error);
4111 }
4112 
4113 /*
4114  * Write out a single page, possibly klustering adjacent dirty pages.
4115  */
4116 int
4117 nfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
4118 	int flags, cred_t *cr)
4119 {
4120 	u_offset_t io_off;
4121 	u_offset_t lbn_off;
4122 	u_offset_t lbn;
4123 	size_t io_len;
4124 	uint_t bsize;
4125 	int error;
4126 	rnode_t *rp;
4127 
4128 	ASSERT(!vn_is_readonly(vp));
4129 	ASSERT(pp != NULL);
4130 	ASSERT(cr != NULL);
4131 	ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone);
4132 
4133 	rp = VTOR(vp);
4134 	ASSERT(rp->r_count > 0);
4135 
4136 	ASSERT(pp->p_offset <= MAXOFF32_T);
4137 
4138 	bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
4139 	lbn = pp->p_offset / bsize;
4140 	lbn_off = lbn * bsize;
4141 
4142 	/*
4143 	 * Find a kluster that fits in one block, or in
4144 	 * one page if pages are bigger than blocks.  If
4145 	 * there is less file space allocated than a whole
4146 	 * page, we'll shorten the i/o request below.
4147 	 */
4148 	pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off,
4149 	    roundup(bsize, PAGESIZE), flags);
4150 
4151 	/*
4152 	 * pvn_write_kluster shouldn't have returned a page with offset
4153 	 * behind the original page we were given.  Verify that.
4154 	 */
4155 	ASSERT((pp->p_offset / bsize) >= lbn);
4156 
4157 	/*
4158 	 * Now pp will have the list of kept dirty pages marked for
4159 	 * write back.  It will also handle invalidation and freeing
4160 	 * of pages that are not dirty.  Check for page length rounding
4161 	 * problems.
4162 	 */
4163 	if (io_off + io_len > lbn_off + bsize) {
4164 		ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE);
4165 		io_len = lbn_off + bsize - io_off;
4166 	}
4167 	/*
4168 	 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4169 	 * consistent value of r_size. RMODINPROGRESS is set in writerp().
4170 	 * When RMODINPROGRESS is set it indicates that a uiomove() is in
4171 	 * progress and the r_size has not been made consistent with the
4172 	 * new size of the file. When the uiomove() completes the r_size is
4173 	 * updated and the RMODINPROGRESS flag is cleared.
4174 	 *
4175 	 * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4176 	 * consistent value of r_size. Without this handshaking, it is
4177 	 * possible that nfs(3)_bio() picks  up the old value of r_size
4178 	 * before the uiomove() in writerp() completes. This will result
4179 	 * in the write through nfs(3)_bio() being dropped.
4180 	 *
4181 	 * More precisely, there is a window between the time the uiomove()
4182 	 * completes and the time the r_size is updated. If a VOP_PUTPAGE()
4183 	 * operation intervenes in this window, the page will be picked up,
4184 	 * because it is dirty (it will be unlocked, unless it was
4185 	 * pagecreate'd). When the page is picked up as dirty, the dirty
4186 	 * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is
4187 	 * checked. This will still be the old size. Therefore the page will
4188 	 * not be written out. When segmap_release() calls VOP_PUTPAGE(),
4189 	 * the page will be found to be clean and the write will be dropped.
4190 	 */
4191 	if (rp->r_flags & RMODINPROGRESS) {
4192 		mutex_enter(&rp->r_statelock);
4193 		if ((rp->r_flags & RMODINPROGRESS) &&
4194 		    rp->r_modaddr + MAXBSIZE > io_off &&
4195 		    rp->r_modaddr < io_off + io_len) {
4196 			page_t *plist;
4197 			/*
4198 			 * A write is in progress for this region of the file.
4199 			 * If we did not detect RMODINPROGRESS here then this
4200 			 * path through nfs_putapage() would eventually go to
4201 			 * nfs(3)_bio() and may not write out all of the data
4202 			 * in the pages. We end up losing data. So we decide
4203 			 * to set the modified bit on each page in the page
4204 			 * list and mark the rnode with RDIRTY. This write
4205 			 * will be restarted at some later time.
4206 			 */
4207 			plist = pp;
4208 			while (plist != NULL) {
4209 				pp = plist;
4210 				page_sub(&plist, pp);
4211 				hat_setmod(pp);
4212 				page_io_unlock(pp);
4213 				page_unlock(pp);
4214 			}
4215 			rp->r_flags |= RDIRTY;
4216 			mutex_exit(&rp->r_statelock);
4217 			if (offp)
4218 				*offp = io_off;
4219 			if (lenp)
4220 				*lenp = io_len;
4221 			return (0);
4222 		}
4223 		mutex_exit(&rp->r_statelock);
4224 	}
4225 
4226 	if (flags & B_ASYNC) {
4227 		error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr,
4228 		    nfs_sync_putapage);
4229 	} else
4230 		error = nfs_sync_putapage(vp, pp, io_off, io_len, flags, cr);
4231 
4232 	if (offp)
4233 		*offp = io_off;
4234 	if (lenp)
4235 		*lenp = io_len;
4236 	return (error);
4237 }
4238 
4239 static int
4240 nfs_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4241 	int flags, cred_t *cr)
4242 {
4243 	int error;
4244 	rnode_t *rp;
4245 
4246 	flags |= B_WRITE;
4247 
4248 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4249 	error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4250 
4251 	rp = VTOR(vp);
4252 
4253 	if ((error == ENOSPC || error == EDQUOT || error == EACCES) &&
4254 	    (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) {
4255 		if (!(rp->r_flags & ROUTOFSPACE)) {
4256 			mutex_enter(&rp->r_statelock);
4257 			rp->r_flags |= ROUTOFSPACE;
4258 			mutex_exit(&rp->r_statelock);
4259 		}
4260 		flags |= B_ERROR;
4261 		pvn_write_done(pp, flags);
4262 		/*
4263 		 * If this was not an async thread, then try again to
4264 		 * write out the pages, but this time, also destroy
4265 		 * them whether or not the write is successful.  This
4266 		 * will prevent memory from filling up with these
4267 		 * pages and destroying them is the only alternative
4268 		 * if they can't be written out.
4269 		 *
4270 		 * Don't do this if this is an async thread because
4271 		 * when the pages are unlocked in pvn_write_done,
4272 		 * some other thread could have come along, locked
4273 		 * them, and queued for an async thread.  It would be
4274 		 * possible for all of the async threads to be tied
4275 		 * up waiting to lock the pages again and they would
4276 		 * all already be locked and waiting for an async
4277 		 * thread to handle them.  Deadlock.
4278 		 */
4279 		if (!(flags & B_ASYNC)) {
4280 			error = nfs_putpage(vp, io_off, io_len,
4281 			    B_INVAL | B_FORCE, cr, NULL);
4282 		}
4283 	} else {
4284 		if (error)
4285 			flags |= B_ERROR;
4286 		else if (rp->r_flags & ROUTOFSPACE) {
4287 			mutex_enter(&rp->r_statelock);
4288 			rp->r_flags &= ~ROUTOFSPACE;
4289 			mutex_exit(&rp->r_statelock);
4290 		}
4291 		pvn_write_done(pp, flags);
4292 	}
4293 
4294 	return (error);
4295 }
4296 
4297 /* ARGSUSED */
4298 static int
4299 nfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4300 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4301 	caller_context_t *ct)
4302 {
4303 	struct segvn_crargs vn_a;
4304 	int error;
4305 	rnode_t *rp;
4306 	struct vattr va;
4307 
4308 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4309 		return (EIO);
4310 
4311 	if (vp->v_flag & VNOMAP)
4312 		return (ENOSYS);
4313 
4314 	if (off > MAXOFF32_T)
4315 		return (EFBIG);
4316 
4317 	if (off < 0 || off + len < 0)
4318 		return (ENXIO);
4319 
4320 	if (vp->v_type != VREG)
4321 		return (ENODEV);
4322 
4323 	/*
4324 	 * If there is cached data and if close-to-open consistency
4325 	 * checking is not turned off and if the file system is not
4326 	 * mounted readonly, then force an over the wire getattr.
4327 	 * Otherwise, just invoke nfsgetattr to get a copy of the
4328 	 * attributes.  The attribute cache will be used unless it
4329 	 * is timed out and if it is, then an over the wire getattr
4330 	 * will be issued.
4331 	 */
4332 	va.va_mask = AT_ALL;
4333 	if (vn_has_cached_data(vp) &&
4334 	    !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp))
4335 		error = nfs_getattr_otw(vp, &va, cr);
4336 	else
4337 		error = nfsgetattr(vp, &va, cr);
4338 	if (error)
4339 		return (error);
4340 
4341 	/*
4342 	 * Check to see if the vnode is currently marked as not cachable.
4343 	 * This means portions of the file are locked (through VOP_FRLOCK).
4344 	 * In this case the map request must be refused.  We use
4345 	 * rp->r_lkserlock to avoid a race with concurrent lock requests.
4346 	 */
4347 	rp = VTOR(vp);
4348 
4349 	/*
4350 	 * Atomically increment r_inmap after acquiring r_rwlock. The
4351 	 * idea here is to acquire r_rwlock to block read/write and
4352 	 * not to protect r_inmap. r_inmap will inform nfs_read/write()
4353 	 * that we are in nfs_map(). Now, r_rwlock is acquired in order
4354 	 * and we can prevent the deadlock that would have occurred
4355 	 * when nfs_addmap() would have acquired it out of order.
4356 	 *
4357 	 * Since we are not protecting r_inmap by any lock, we do not
4358 	 * hold any lock when we decrement it. We atomically decrement
4359 	 * r_inmap after we release r_lkserlock.
4360 	 */
4361 
4362 	if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp)))
4363 		return (EINTR);
4364 	atomic_inc_uint(&rp->r_inmap);
4365 	nfs_rw_exit(&rp->r_rwlock);
4366 
4367 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) {
4368 		atomic_dec_uint(&rp->r_inmap);
4369 		return (EINTR);
4370 	}
4371 	if (vp->v_flag & VNOCACHE) {
4372 		error = EAGAIN;
4373 		goto done;
4374 	}
4375 
4376 	/*
4377 	 * Don't allow concurrent locks and mapping if mandatory locking is
4378 	 * enabled.
4379 	 */
4380 	if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) &&
4381 	    MANDLOCK(vp, va.va_mode)) {
4382 		error = EAGAIN;
4383 		goto done;
4384 	}
4385 
4386 	as_rangelock(as);
4387 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4388 	if (error != 0) {
4389 		as_rangeunlock(as);
4390 		goto done;
4391 	}
4392 
4393 	vn_a.vp = vp;
4394 	vn_a.offset = off;
4395 	vn_a.type = (flags & MAP_TYPE);
4396 	vn_a.prot = (uchar_t)prot;
4397 	vn_a.maxprot = (uchar_t)maxprot;
4398 	vn_a.flags = (flags & ~MAP_TYPE);
4399 	vn_a.cred = cr;
4400 	vn_a.amp = NULL;
4401 	vn_a.szc = 0;
4402 	vn_a.lgrp_mem_policy_flags = 0;
4403 
4404 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4405 	as_rangeunlock(as);
4406 
4407 done:
4408 	nfs_rw_exit(&rp->r_lkserlock);
4409 	atomic_dec_uint(&rp->r_inmap);
4410 	return (error);
4411 }
4412 
4413 /* ARGSUSED */
4414 static int
4415 nfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4416 	size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4417 	caller_context_t *ct)
4418 {
4419 	rnode_t *rp;
4420 
4421 	if (vp->v_flag & VNOMAP)
4422 		return (ENOSYS);
4423 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4424 		return (EIO);
4425 
4426 	rp = VTOR(vp);
4427 	atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len));
4428 
4429 	return (0);
4430 }
4431 
4432 /* ARGSUSED */
4433 static int
4434 nfs_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, offset_t offset,
4435 	struct flk_callback *flk_cbp, cred_t *cr, caller_context_t *ct)
4436 {
4437 	netobj lm_fh;
4438 	int rc;
4439 	u_offset_t start, end;
4440 	rnode_t *rp;
4441 	int error = 0, intr = INTR(vp);
4442 
4443 	/* check for valid cmd parameter */
4444 	if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW)
4445 		return (EINVAL);
4446 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4447 		return (EIO);
4448 
4449 	/* Verify l_type. */
4450 	switch (bfp->l_type) {
4451 	case F_RDLCK:
4452 		if (cmd != F_GETLK && !(flag & FREAD))
4453 			return (EBADF);
4454 		break;
4455 	case F_WRLCK:
4456 		if (cmd != F_GETLK && !(flag & FWRITE))
4457 			return (EBADF);
4458 		break;
4459 	case F_UNLCK:
4460 		intr = 0;
4461 		break;
4462 
4463 	default:
4464 		return (EINVAL);
4465 	}
4466 
4467 	/* check the validity of the lock range */
4468 	if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset))
4469 		return (rc);
4470 	if (rc = flk_check_lock_data(start, end, MAXOFF32_T))
4471 		return (rc);
4472 
4473 	/*
4474 	 * If the filesystem is mounted using local locking, pass the
4475 	 * request off to the local locking code.
4476 	 */
4477 	if (VTOMI(vp)->mi_flags & MI_LLOCK) {
4478 		if (offset > MAXOFF32_T)
4479 			return (EFBIG);
4480 		if (cmd == F_SETLK || cmd == F_SETLKW) {
4481 			/*
4482 			 * For complete safety, we should be holding
4483 			 * r_lkserlock.  However, we can't call
4484 			 * lm_safelock and then fs_frlock while
4485 			 * holding r_lkserlock, so just invoke
4486 			 * lm_safelock and expect that this will
4487 			 * catch enough of the cases.
4488 			 */
4489 			if (!lm_safelock(vp, bfp, cr))
4490 				return (EAGAIN);
4491 		}
4492 		return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4493 	}
4494 
4495 	rp = VTOR(vp);
4496 
4497 	/*
4498 	 * Check whether the given lock request can proceed, given the
4499 	 * current file mappings.
4500 	 */
4501 	if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr))
4502 		return (EINTR);
4503 	if (cmd == F_SETLK || cmd == F_SETLKW) {
4504 		if (!lm_safelock(vp, bfp, cr)) {
4505 			rc = EAGAIN;
4506 			goto done;
4507 		}
4508 	}
4509 
4510 	/*
4511 	 * Flush the cache after waiting for async I/O to finish.  For new
4512 	 * locks, this is so that the process gets the latest bits from the
4513 	 * server.  For unlocks, this is so that other clients see the
4514 	 * latest bits once the file has been unlocked.  If currently dirty
4515 	 * pages can't be flushed, then don't allow a lock to be set.  But
4516 	 * allow unlocks to succeed, to avoid having orphan locks on the
4517 	 * server.
4518 	 */
4519 	if (cmd != F_GETLK) {
4520 		mutex_enter(&rp->r_statelock);
4521 		while (rp->r_count > 0) {
4522 			if (intr) {
4523 				klwp_t *lwp = ttolwp(curthread);
4524 
4525 				if (lwp != NULL)
4526 					lwp->lwp_nostop++;
4527 				if (cv_wait_sig(&rp->r_cv, &rp->r_statelock)
4528 				    == 0) {
4529 					if (lwp != NULL)
4530 						lwp->lwp_nostop--;
4531 					rc = EINTR;
4532 					break;
4533 				}
4534 				if (lwp != NULL)
4535 					lwp->lwp_nostop--;
4536 			} else
4537 			cv_wait(&rp->r_cv, &rp->r_statelock);
4538 		}
4539 		mutex_exit(&rp->r_statelock);
4540 		if (rc != 0)
4541 			goto done;
4542 		error = nfs_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct);
4543 		if (error) {
4544 			if (error == ENOSPC || error == EDQUOT) {
4545 				mutex_enter(&rp->r_statelock);
4546 				if (!rp->r_error)
4547 					rp->r_error = error;
4548 				mutex_exit(&rp->r_statelock);
4549 			}
4550 			if (bfp->l_type != F_UNLCK) {
4551 				rc = ENOLCK;
4552 				goto done;
4553 			}
4554 		}
4555 	}
4556 
4557 	lm_fh.n_len = sizeof (fhandle_t);
4558 	lm_fh.n_bytes = (char *)VTOFH(vp);
4559 
4560 	/*
4561 	 * Call the lock manager to do the real work of contacting
4562 	 * the server and obtaining the lock.
4563 	 */
4564 	rc = lm_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh, flk_cbp);
4565 
4566 	if (rc == 0)
4567 		nfs_lockcompletion(vp, cmd);
4568 
4569 done:
4570 	nfs_rw_exit(&rp->r_lkserlock);
4571 	return (rc);
4572 }
4573 
4574 /*
4575  * Free storage space associated with the specified vnode.  The portion
4576  * to be freed is specified by bfp->l_start and bfp->l_len (already
4577  * normalized to a "whence" of 0).
4578  *
4579  * This is an experimental facility whose continued existence is not
4580  * guaranteed.  Currently, we only support the special case
4581  * of l_len == 0, meaning free to end of file.
4582  */
4583 /* ARGSUSED */
4584 static int
4585 nfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
4586 	offset_t offset, cred_t *cr, caller_context_t *ct)
4587 {
4588 	int error;
4589 
4590 	ASSERT(vp->v_type == VREG);
4591 	if (cmd != F_FREESP)
4592 		return (EINVAL);
4593 
4594 	if (offset > MAXOFF32_T)
4595 		return (EFBIG);
4596 
4597 	if ((bfp->l_start > MAXOFF32_T) || (bfp->l_end > MAXOFF32_T) ||
4598 	    (bfp->l_len > MAXOFF32_T))
4599 		return (EFBIG);
4600 
4601 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4602 		return (EIO);
4603 
4604 	error = convoff(vp, bfp, 0, offset);
4605 	if (!error) {
4606 		ASSERT(bfp->l_start >= 0);
4607 		if (bfp->l_len == 0) {
4608 			struct vattr va;
4609 
4610 			/*
4611 			 * ftruncate should not change the ctime and
4612 			 * mtime if we truncate the file to its
4613 			 * previous size.
4614 			 */
4615 			va.va_mask = AT_SIZE;
4616 			error = nfsgetattr(vp, &va, cr);
4617 			if (error || va.va_size == bfp->l_start)
4618 				return (error);
4619 			va.va_mask = AT_SIZE;
4620 			va.va_size = bfp->l_start;
4621 			error = nfssetattr(vp, &va, 0, cr);
4622 
4623 			if (error == 0 && bfp->l_start == 0)
4624 				vnevent_truncate(vp, ct);
4625 		} else
4626 			error = EINVAL;
4627 	}
4628 
4629 	return (error);
4630 }
4631 
4632 /* ARGSUSED */
4633 static int
4634 nfs_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
4635 {
4636 
4637 	return (EINVAL);
4638 }
4639 
4640 /*
4641  * Setup and add an address space callback to do the work of the delmap call.
4642  * The callback will (and must be) deleted in the actual callback function.
4643  *
4644  * This is done in order to take care of the problem that we have with holding
4645  * the address space's a_lock for a long period of time (e.g. if the NFS server
4646  * is down).  Callbacks will be executed in the address space code while the
4647  * a_lock is not held.	Holding the address space's a_lock causes things such
4648  * as ps and fork to hang because they are trying to acquire this lock as well.
4649  */
4650 /* ARGSUSED */
4651 static int
4652 nfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4653 	size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4654 	caller_context_t *ct)
4655 {
4656 	int			caller_found;
4657 	int			error;
4658 	rnode_t			*rp;
4659 	nfs_delmap_args_t	*dmapp;
4660 	nfs_delmapcall_t	*delmap_call;
4661 
4662 	if (vp->v_flag & VNOMAP)
4663 		return (ENOSYS);
4664 	/*
4665 	 * A process may not change zones if it has NFS pages mmap'ed
4666 	 * in, so we can't legitimately get here from the wrong zone.
4667 	 */
4668 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4669 
4670 	rp = VTOR(vp);
4671 
4672 	/*
4673 	 * The way that the address space of this process deletes its mapping
4674 	 * of this file is via the following call chains:
4675 	 * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4676 	 * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4677 	 *
4678 	 * With the use of address space callbacks we are allowed to drop the
4679 	 * address space lock, a_lock, while executing the NFS operations that
4680 	 * need to go over the wire.  Returning EAGAIN to the caller of this
4681 	 * function is what drives the execution of the callback that we add
4682 	 * below.  The callback will be executed by the address space code
4683 	 * after dropping the a_lock.  When the callback is finished, since
4684 	 * we dropped the a_lock, it must be re-acquired and segvn_unmap()
4685 	 * is called again on the same segment to finish the rest of the work
4686 	 * that needs to happen during unmapping.
4687 	 *
4688 	 * This action of calling back into the segment driver causes
4689 	 * nfs_delmap() to get called again, but since the callback was
4690 	 * already executed at this point, it already did the work and there
4691 	 * is nothing left for us to do.
4692 	 *
4693 	 * To Summarize:
4694 	 * - The first time nfs_delmap is called by the current thread is when
4695 	 * we add the caller associated with this delmap to the delmap caller
4696 	 * list, add the callback, and return EAGAIN.
4697 	 * - The second time in this call chain when nfs_delmap is called we
4698 	 * will find this caller in the delmap caller list and realize there
4699 	 * is no more work to do thus removing this caller from the list and
4700 	 * returning the error that was set in the callback execution.
4701 	 */
4702 	caller_found = nfs_find_and_delete_delmapcall(rp, &error);
4703 	if (caller_found) {
4704 		/*
4705 		 * 'error' is from the actual delmap operations.  To avoid
4706 		 * hangs, we need to handle the return of EAGAIN differently
4707 		 * since this is what drives the callback execution.
4708 		 * In this case, we don't want to return EAGAIN and do the
4709 		 * callback execution because there are none to execute.
4710 		 */
4711 		if (error == EAGAIN)
4712 			return (0);
4713 		else
4714 			return (error);
4715 	}
4716 
4717 	/* current caller was not in the list */
4718 	delmap_call = nfs_init_delmapcall();
4719 
4720 	mutex_enter(&rp->r_statelock);
4721 	list_insert_tail(&rp->r_indelmap, delmap_call);
4722 	mutex_exit(&rp->r_statelock);
4723 
4724 	dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP);
4725 
4726 	dmapp->vp = vp;
4727 	dmapp->off = off;
4728 	dmapp->addr = addr;
4729 	dmapp->len = len;
4730 	dmapp->prot = prot;
4731 	dmapp->maxprot = maxprot;
4732 	dmapp->flags = flags;
4733 	dmapp->cr = cr;
4734 	dmapp->caller = delmap_call;
4735 
4736 	error = as_add_callback(as, nfs_delmap_callback, dmapp,
4737 	    AS_UNMAP_EVENT, addr, len, KM_SLEEP);
4738 
4739 	return (error ? error : EAGAIN);
4740 }
4741 
4742 /*
4743  * Remove some pages from an mmap'd vnode.  Just update the
4744  * count of pages.  If doing close-to-open, then flush all
4745  * of the pages associated with this file.  Otherwise, start
4746  * an asynchronous page flush to write out any dirty pages.
4747  * This will also associate a credential with the rnode which
4748  * can be used to write the pages.
4749  */
4750 /* ARGSUSED */
4751 static void
4752 nfs_delmap_callback(struct as *as, void *arg, uint_t event)
4753 {
4754 	int			error;
4755 	rnode_t			*rp;
4756 	mntinfo_t		*mi;
4757 	nfs_delmap_args_t	*dmapp = (nfs_delmap_args_t *)arg;
4758 
4759 	rp = VTOR(dmapp->vp);
4760 	mi = VTOMI(dmapp->vp);
4761 
4762 	atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len));
4763 	ASSERT(rp->r_mapcnt >= 0);
4764 
4765 	/*
4766 	 * Initiate a page flush if there are pages, the file system
4767 	 * was not mounted readonly, the segment was mapped shared, and
4768 	 * the pages themselves were writeable.
4769 	 */
4770 	if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) &&
4771 	    dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) {
4772 		mutex_enter(&rp->r_statelock);
4773 		rp->r_flags |= RDIRTY;
4774 		mutex_exit(&rp->r_statelock);
4775 		/*
4776 		 * If this is a cross-zone access a sync putpage won't work, so
4777 		 * the best we can do is try an async putpage.  That seems
4778 		 * better than something more draconian such as discarding the
4779 		 * dirty pages.
4780 		 */
4781 		if ((mi->mi_flags & MI_NOCTO) ||
4782 		    nfs_zone() != mi->mi_zone)
4783 			error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4784 			    B_ASYNC, dmapp->cr, NULL);
4785 		else
4786 			error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4787 			    0, dmapp->cr, NULL);
4788 		if (!error) {
4789 			mutex_enter(&rp->r_statelock);
4790 			error = rp->r_error;
4791 			rp->r_error = 0;
4792 			mutex_exit(&rp->r_statelock);
4793 		}
4794 	} else
4795 		error = 0;
4796 
4797 	if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO))
4798 		(void) nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4799 		    B_INVAL, dmapp->cr, NULL);
4800 
4801 	dmapp->caller->error = error;
4802 	(void) as_delete_callback(as, arg);
4803 	kmem_free(dmapp, sizeof (nfs_delmap_args_t));
4804 }
4805 
4806 /* ARGSUSED */
4807 static int
4808 nfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4809 	caller_context_t *ct)
4810 {
4811 	int error = 0;
4812 
4813 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4814 		return (EIO);
4815 	/*
4816 	 * This looks a little weird because it's written in a general
4817 	 * manner but we make little use of cases.  If cntl() ever gets
4818 	 * widely used, the outer switch will make more sense.
4819 	 */
4820 
4821 	switch (cmd) {
4822 
4823 	/*
4824 	 * Large file spec - need to base answer new query with
4825 	 * hardcoded constant based on the protocol.
4826 	 */
4827 	case _PC_FILESIZEBITS:
4828 		*valp = 32;
4829 		return (0);
4830 
4831 	case _PC_LINK_MAX:
4832 	case _PC_NAME_MAX:
4833 	case _PC_PATH_MAX:
4834 	case _PC_SYMLINK_MAX:
4835 	case _PC_CHOWN_RESTRICTED:
4836 	case _PC_NO_TRUNC: {
4837 		mntinfo_t *mi;
4838 		struct pathcnf *pc;
4839 
4840 		if ((mi = VTOMI(vp)) == NULL || (pc = mi->mi_pathconf) == NULL)
4841 			return (EINVAL);
4842 		error = _PC_ISSET(cmd, pc->pc_mask);    /* error or bool */
4843 		switch (cmd) {
4844 		case _PC_LINK_MAX:
4845 			*valp = pc->pc_link_max;
4846 			break;
4847 		case _PC_NAME_MAX:
4848 			*valp = pc->pc_name_max;
4849 			break;
4850 		case _PC_PATH_MAX:
4851 		case _PC_SYMLINK_MAX:
4852 			*valp = pc->pc_path_max;
4853 			break;
4854 		case _PC_CHOWN_RESTRICTED:
4855 			/*
4856 			 * if we got here, error is really a boolean which
4857 			 * indicates whether cmd is set or not.
4858 			 */
4859 			*valp = error ? 1 : 0;	/* see above */
4860 			error = 0;
4861 			break;
4862 		case _PC_NO_TRUNC:
4863 			/*
4864 			 * if we got here, error is really a boolean which
4865 			 * indicates whether cmd is set or not.
4866 			 */
4867 			*valp = error ? 1 : 0;	/* see above */
4868 			error = 0;
4869 			break;
4870 		}
4871 		return (error ? EINVAL : 0);
4872 		}
4873 
4874 	case _PC_XATTR_EXISTS:
4875 		*valp = 0;
4876 		if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
4877 			vnode_t *avp;
4878 			rnode_t *rp;
4879 			mntinfo_t *mi = VTOMI(vp);
4880 
4881 			if (!(mi->mi_flags & MI_EXTATTR))
4882 				return (0);
4883 
4884 			rp = VTOR(vp);
4885 			if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER,
4886 			    INTR(vp)))
4887 				return (EINTR);
4888 
4889 			error = nfslookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr);
4890 			if (error || avp == NULL)
4891 				error = acl_getxattrdir2(vp, &avp, 0, cr, 0);
4892 
4893 			nfs_rw_exit(&rp->r_rwlock);
4894 
4895 			if (error == 0 && avp != NULL) {
4896 				error = do_xattr_exists_check(avp, valp, cr);
4897 				VN_RELE(avp);
4898 			}
4899 		}
4900 		return (error ? EINVAL : 0);
4901 
4902 	case _PC_ACL_ENABLED:
4903 		*valp = _ACL_ACLENT_ENABLED;
4904 		return (0);
4905 
4906 	default:
4907 		return (EINVAL);
4908 	}
4909 }
4910 
4911 /*
4912  * Called by async thread to do synchronous pageio. Do the i/o, wait
4913  * for it to complete, and cleanup the page list when done.
4914  */
4915 static int
4916 nfs_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4917 	int flags, cred_t *cr)
4918 {
4919 	int error;
4920 
4921 	ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4922 	error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4923 	if (flags & B_READ)
4924 		pvn_read_done(pp, (error ? B_ERROR : 0) | flags);
4925 	else
4926 		pvn_write_done(pp, (error ? B_ERROR : 0) | flags);
4927 	return (error);
4928 }
4929 
4930 /* ARGSUSED */
4931 static int
4932 nfs_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4933 	int flags, cred_t *cr, caller_context_t *ct)
4934 {
4935 	int error;
4936 	rnode_t *rp;
4937 
4938 	if (pp == NULL)
4939 		return (EINVAL);
4940 
4941 	if (io_off > MAXOFF32_T)
4942 		return (EFBIG);
4943 	if (nfs_zone() != VTOMI(vp)->mi_zone)
4944 		return (EIO);
4945 	rp = VTOR(vp);
4946 	mutex_enter(&rp->r_statelock);
4947 	rp->r_count++;
4948 	mutex_exit(&rp->r_statelock);
4949 
4950 	if (flags & B_ASYNC) {
4951 		error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr,
4952 		    nfs_sync_pageio);
4953 	} else
4954 		error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4955 	mutex_enter(&rp->r_statelock);
4956 	rp->r_count--;
4957 	cv_broadcast(&rp->r_cv);
4958 	mutex_exit(&rp->r_statelock);
4959 	return (error);
4960 }
4961 
4962 /* ARGSUSED */
4963 static int
4964 nfs_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4965 	caller_context_t *ct)
4966 {
4967 	int error;
4968 	mntinfo_t *mi;
4969 
4970 	mi = VTOMI(vp);
4971 
4972 	if (nfs_zone() != mi->mi_zone)
4973 		return (EIO);
4974 	if (mi->mi_flags & MI_ACL) {
4975 		error = acl_setacl2(vp, vsecattr, flag, cr);
4976 		if (mi->mi_flags & MI_ACL)
4977 			return (error);
4978 	}
4979 
4980 	return (ENOSYS);
4981 }
4982 
4983 /* ARGSUSED */
4984 static int
4985 nfs_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4986 	caller_context_t *ct)
4987 {
4988 	int error;
4989 	mntinfo_t *mi;
4990 
4991 	mi = VTOMI(vp);
4992 
4993 	if (nfs_zone() != mi->mi_zone)
4994 		return (EIO);
4995 	if (mi->mi_flags & MI_ACL) {
4996 		error = acl_getacl2(vp, vsecattr, flag, cr);
4997 		if (mi->mi_flags & MI_ACL)
4998 			return (error);
4999 	}
5000 
5001 	return (fs_fab_acl(vp, vsecattr, flag, cr, ct));
5002 }
5003 
5004 /* ARGSUSED */
5005 static int
5006 nfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
5007 	caller_context_t *ct)
5008 {
5009 	int error;
5010 	struct shrlock nshr;
5011 	struct nfs_owner nfs_owner;
5012 	netobj lm_fh;
5013 
5014 	if (nfs_zone() != VTOMI(vp)->mi_zone)
5015 		return (EIO);
5016 
5017 	/*
5018 	 * check for valid cmd parameter
5019 	 */
5020 	if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS)
5021 		return (EINVAL);
5022 
5023 	/*
5024 	 * Check access permissions
5025 	 */
5026 	if (cmd == F_SHARE &&
5027 	    (((shr->s_access & F_RDACC) && !(flag & FREAD)) ||
5028 	    ((shr->s_access & F_WRACC) && !(flag & FWRITE))))
5029 		return (EBADF);
5030 
5031 	/*
5032 	 * If the filesystem is mounted using local locking, pass the
5033 	 * request off to the local share code.
5034 	 */
5035 	if (VTOMI(vp)->mi_flags & MI_LLOCK)
5036 		return (fs_shrlock(vp, cmd, shr, flag, cr, ct));
5037 
5038 	switch (cmd) {
5039 	case F_SHARE:
5040 	case F_UNSHARE:
5041 		lm_fh.n_len = sizeof (fhandle_t);
5042 		lm_fh.n_bytes = (char *)VTOFH(vp);
5043 
5044 		/*
5045 		 * If passed an owner that is too large to fit in an
5046 		 * nfs_owner it is likely a recursive call from the
5047 		 * lock manager client and pass it straight through.  If
5048 		 * it is not a nfs_owner then simply return an error.
5049 		 */
5050 		if (shr->s_own_len > sizeof (nfs_owner.lowner)) {
5051 			if (((struct nfs_owner *)shr->s_owner)->magic !=
5052 			    NFS_OWNER_MAGIC)
5053 				return (EINVAL);
5054 
5055 			if (error = lm_shrlock(vp, cmd, shr, flag, &lm_fh)) {
5056 				error = set_errno(error);
5057 			}
5058 			return (error);
5059 		}
5060 		/*
5061 		 * Remote share reservations owner is a combination of
5062 		 * a magic number, hostname, and the local owner
5063 		 */
5064 		bzero(&nfs_owner, sizeof (nfs_owner));
5065 		nfs_owner.magic = NFS_OWNER_MAGIC;
5066 		(void) strncpy(nfs_owner.hname, uts_nodename(),
5067 		    sizeof (nfs_owner.hname));
5068 		bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len);
5069 		nshr.s_access = shr->s_access;
5070 		nshr.s_deny = shr->s_deny;
5071 		nshr.s_sysid = 0;
5072 		nshr.s_pid = ttoproc(curthread)->p_pid;
5073 		nshr.s_own_len = sizeof (nfs_owner);
5074 		nshr.s_owner = (caddr_t)&nfs_owner;
5075 
5076 		if (error = lm_shrlock(vp, cmd, &nshr, flag, &lm_fh)) {
5077 			error = set_errno(error);
5078 		}
5079 
5080 		break;
5081 
5082 	case F_HASREMOTELOCKS:
5083 		/*
5084 		 * NFS client can't store remote locks itself
5085 		 */
5086 		shr->s_access = 0;
5087 		error = 0;
5088 		break;
5089 
5090 	default:
5091 		error = EINVAL;
5092 		break;
5093 	}
5094 
5095 	return (error);
5096 }
5097